Science Technology Engineering Math (STEM)
Classes and Females' Career Choices
Chapter 1: Introduction to the Study
Introduction
In management, there are several functions. These basic functions
include planning, organizing, staffing, controlling, evaluating, and
implementing. In planning a career, many of these functions of management
integrate with one another (Drucker, 1954). There is also management by
objective (MBO), which is the importance of setting and accomplishing
clear goals and objectives closely related to the concept of planning such as
in the field of career management (Drucker, 1954; Seibert, Kraimer, Holtom,
& Pierotti, 2013). In this context, the objectives dealt with the field of career
management as it applies to career choices for females, particularly in
science, technology, engineering, and math (STEM),
One aspect of career management that I focused on is helping females
manage their careers by encouraging them to choose STEM careers to close
the gender gap (Hensvik, 2014). Hence, it is important to encourage more
females in STEM careers to increase diversity of ideas in these work places,
reducing the gender gap in these fields and encouraging teamwork among
males and females (Seibert et al., 2013; Senge, 2006). When a young female
has a plan for a STEM career, she needs to set goals, objectives, and
evaluate the classes that she needs to take in high school and postsecondary
school to accomplish her objectives to manage her career. Moreover, she
must set career goals and objectives at each step beginning with her
education. She also needs to be organized in order to manage her career
plans efficiently and effectively and access these high-paying fields.
Encouraging more females to plan their careers in STEM can reduce the
gender gap and increase diversity of ideas in these workplaces (Hensvik,
2014).
There is still a clear gender gap in STEM careers. According to the
Department of Commerce (2011), less than 24% of STEM positions in the
United States are occupied by females. Despite their success filling male’s
jobs during WWII and throughout history, today there are only 20% females
in STEM careers, with females making up 47% of the workforce (Kenney,
McGee, & Bhatnagar, 2012). Although the place of females in society has
changed, many still believe a female’s place is in the home (Kellerman &
Rhode, 2007; Kenney et al., 2012). This kind of attitude contributes to this
gender gap because many believe that females do not belong in the sciences
(Farland-Smith, 2009; He & Freeman, 2010: Kenney et al., 2012; Sikora &
Pokropek, 2012). Encouraging females to become interested in math and
science at a young age can bring about social change by reducing the gender
gap in these fields, helping females to access higherpaying careers (Hensvik,
2014).
Many factors such as not being encouraged in the sciences
academically, the lack of same-sex STEM role models, and thoughts that
females are not as good in STEM as males have discouraged these females
from planning for STEM careers, since childhood (Farland-Smith, 2009; He
& Freeman, 2010, Sikora & Pokropek, 2012) because females may not be
encouraged to take as many STEM classes as males and may not have the
confidence in their abilities to perform these mathematical and scientific
tasks. Consequently, they may take fewer STEM classes and have fewer
same sex role models than their male counterparts. The play activities
females partake in as young children from preschool to preteen can have an
impact because engaging in science activities as youth increases an interest
in science, which may increase the chance of choosing a STEM career
(Noddings, 1986). To educate on this gender gap, I conducted a quantitative
study where I investigated the relationship between the number of STEM
classes that females in my sample born after 1980 have taken in high school
and college, how many same sex role models they had to encourage them,
and the relationship to their career choices and salaries. My hypothesis was
that the more STEM classes females take during their younger years,
especially in high school and postsecondary school, and the more same sex
role models that they have had to encourage them, the more likely they are
to choose a STEM career and receive a high salary.
Background of the Problem
Reasons for Gender Gap
Scholars have suggested numerous reasons for the gender gap in
STEM fields. For example, traditionally, many females were not exposed to
science as children from preschool to preteen. According to Erikson (1980),
prior to the age of 13, females were usually encouraged to play mommy
roles as children (Townsend, 2013) and were not encouraged to play with
scientific or mechanical things. Consequently, many females were not
encouraged to take STEM classes or enter these fields as often as males (He
& Freeman, 2010; Townsend, 2013). Although this trend is changing with
the population I studied, who were born after 1980, it is still evident when
viewing advertising and going into a retail store that many of the playthings
for children are still gender segregated, following traditional gender
stereotypes (Gilligan, 2008). This fact reinforces the gender stereotypes that
females are more suited towards domestic roles and less suited towards
technical roles of math and science, affecting career management. Although
Gilligan (2008) did not specify a percentage, based on personal observation,
it seemed that approximately 50% of playthings are still gender segregated,
reinforcing traditional gender roles. These playthings reinforce girls’ self-
perceptions of their expected roles and traditional gender norms across
society. Furthermore, because of this gender segregation early in life
(Erikson, 1980), and since females are discouraged from taking STEM
classes and entering these fields, there have been views that females are not
as good at math and science as males (He & Freeman, 2010; Stout,
Dasgupta, Hunsinger, & McManus, 2011), affecting career management.
There are other factors contributing to this gender gap. Despite the
perceptions that females are not as good at the sciences as males, they
traditionally outperformed males in math but not in university entrance
exams, in which males fared better. Correll (2004), and Kenney et al, (2012)
compared the differences between the sexes on the abilities of math and
science. They asserted that historically females outperformed males in the
math classes, but the males outperformed the females in high stake
university entrance exams like the SATs, which contributed to why females
avoid STEM careers (Kenney et al, 2012), creating a need for my
relationship study. Like Correll (2004), Kenney et al, (2012) also stated that
females were outperformed in spatial skills because females were made to
feel they were less competent than males in these mathematical, scientific,
technical, and spatial skills in cultures across the board, which stems from
childhood, according to Erikson (1980). As a result of these lower
performances on entrance exams and on spatial skills, according to Correll
(2004), females have had lower self-esteem and a lower opinion of their
math abilities, which hinders them and creates a gender gap, further
discouraging females from entering STEM careers. Therefore, females avoid
these careers because of what was perceived as demanding academics,
which may have contributed to reduced interest in the sciences (Correll;
2004; Kenney et al, 2012) and has widened the gender gap (Kenney et al,
2012). Furthermore, females have not been encouraged to partake in as
many STEM activities, classes, and careers as their male counterparts
(Correll, 2004; Sikora & Pokropek, 2012), due to discrimination and societal
attitudes, which stem from childhood (Erikson, 1980).
Consequently, many females may take less STEM classes for fear of
failure, further contributing to this gender gap. Other factors may be the lack
of female role models, and this confidence gap may have contributed to the
gender gap in STEM careers and low rate of employment due to
underestimation of one’s abilities (Dunning, Kruger, & Williams, 2013;
Farland-Smith, 2009; Glass, Sassler, Levitte, & Michelmore, 2013; Hensvik;
2014; Sikora & Pokropek, 2012). According to the Dunning-Kruger effect,
those who perform well tend to underestimate their abilities, and those who
perform poorly tend to overestimate their abilities (Dunning et al, 2013).
Moreover, males tend to overestimate their abilities, and females tend to
underestimate their abilities, especially in STEM abilities, according to the
research conducted by Kay and Shipman (2014).
Another contribution to the gender gap is the opinion that females
have about sciences being boring and fear being labeled a geek. Thus issue
was brought up by Farland-Smith (2009) and Klawe (2013), the latter from
the famous Harvey-Mudd College for computer science, who postulated that
females were made to feel that science and math were boring, difficult, and
for those labeled geeks. Furthermore, according to Farland-Smith -2009),
there has been a lack of role models for females to emulate in these fields.
Moreover, because of the perception by females that they must be the
primary caregivers in the family, many have avoided STEM careers because
these careers are not as flexible with their hours, making it difficult for
females to balance work and family (Correll, 2004; He & Freeman, 2010).
Another reason for the gender gap, according to Noddings (1986) was
that many females avoided STEM careers for the caring professions,
contributing to the gender gap. This gender gap may occur because females
may be discouraged from taking math and science classes, which could
influence their career choices and salaries, later in life, according to Miller
(2006) and Noddings. Also, according to de Beauvoir (1945), females have
been treated as second class citizens, and this discouragement from these
high paying careers into caring professions is an example of how females
have traditionally been treated like second class citizens (Sherblom, 2008).
The Need for This Study
As a result of these gender segregations stemming from childhood,
and females traditionally being discouraged from math and science, there is
a need for my study. Therefore, this need becomes crucial because I wanted
to find out how many STEM classes the females in my sample have taken
and how many role models they had to determine the impact on their career
choices and salaries to determine how encouraged they were to enter these
fields. To give more rationale on the need for this study, descriptions of the
focus and population are described below.
Despite the feminization of industries like public relations, females
still earn less than males and have fewer opportunities in the STEM fields
and leadership positions than males (Anderson, 2006; Dugan, Fath, Howes,
Lavelle, & Polanin, 2013). Even professions such as accounting are still
dominated by males because of the fear of the financial analysis and math
involved, according to Dambrin and Lambert (2006). Erikson (1971)
recognized this educational connection by stating that the decisions of
classes made in high school and college may impact career choices later on.
I emphasized this using a specific Long Island population, which closes a
gap in the literature. Kenney et al, (2012) believed the gender gap exists
because science careers have been historically associated with masculinity
and not femininity. Females face discrimination in reference to their abilities
in these fields (Kenney et al, 2012). Stout et al, (2011) stated that since
childhood, females have been consistently told that males are better in math
and science than females, which has discouraged females from entering
these fields. However, Stout et al, (2012) concluded that when females saw a
few role models, it increased their self concept in these fields by inoculating
the negative stereotypes about females not being as good in math and
science as males. Stout et al realized that many females had a low self
concept of their math and science abilities. Watt, Shapka, Morris, Durik,
Keating, and Eccles, (2012) postulated that the participation rate for
Australia, Canada, and the United States in STEM careers is the lowest for
females. According to Watt et al, (2012), the reasons for this low
participation rate are the high dropout rates and the restrictive course choices
at the entrance level in a university. These restrictions may deter
participation in these careers, especially for females who tend to avoid
taking a lot of math and science in high school to enter these majors in the
university.
There are several potential benefits from this study. These benefits
include educating on how females can earn higher salaries through entering
STEM careers. Benefits also include helping to make society aware of how
childhood activities can either encourage or discourage females from
entering these careers, and how same-sex role models can encourage these
females to enter these high-paying, competitive careers.
Problem Statement
According to the Department of Commerce (2011), less than 25% of
STEM positions in the United States are occupied by females, creating a
gender gap. Therefore, the principal management problem addressed in this
study is that women as human capital have been discouraged from taking
math and science classes during high school and college, resulting in less
access to these high-paying careers and creating a gender gap, impacting
career management (Brown, Brown, Reardon, & Merrill, 2011; Correll,
2004; Hensvik, 2014; Milgram, 2011). Management consists of planning,
organizing, and controlling, and career management includes these same
elements (Argyris, 1991; Drucker, 1954). According to Drucker (1954) and
Seibert et al, (2013), career management, like all fields of management, is
about setting intrinsic and extrinsic goals. Discrimination, the gender gap,
and discouragement makes it more difficult for females to plan and organize
their careers and meet these goals such as accessing careers in
STEM (Brown et al, 2011; Correll, 2004; He & Freeman, 2010, Hensvik,
2014; Milgram, 2011). According to the Bureau of Labor Statistics (2009),
STEM careers tend to have higher salaries than other career choices.
According to Mavriplis et al, (2010), many females even drop out of STEM
classes. The problem is that females tend to avoid STEM classes in high
school and college, and are thus less likely to have same-sex STEM role
models. These conditions may contribute to why females are less likely to
enter these careers than their male counterparts. This stated relationship was
investigated quantitatively using a survey for data collection and a regression
and a one-way ANOVA for data analysis (Aczel & Sounderpandian, 2009;
Field, 2013), making quantitative data available about this relationship.
The gap in the literature is that I did not find any researchers who
have focused on this gender gap in STEM classes and careers for females
born after 1980, who avoided STEM careers because of the desire for a
flexible career, with less challenging academics (Correll, 2004). Also, few
researchers have examined the gender gap in STEM classes due to a lack of
role models affecting career choices and salaries, which is an aspect of
career management in the workplace (Farland-Smith, 2009); therefore, my
study builds upon prior research conducted in the last 5 years, including
studies by Cornell, FarlandSmith, and Milgram (2011). I examined the effect
of the number of STEM classes taken, the number of role models females
have in high school and postsecondary education, and their relationship to
career choices and salaries.
The lack of same-sex role models, the number of STEM classes that
females take in high school and college, and females’ perceptions of math
and science are important issues to focus on because it is these issues that
may discourage females from highpaying, competitive STEM careers.
Understanding these issues is important in studying my chosen population to
determine which of these factors have either helped or hindered this
population from choosing a STEM career. Furthermore, the connection
between their childhood activities and their chosen career is important
because childhood gender roles may impact females’ perceptions of math
and science later in life (Erikson, 1980).
Purpose of the Study
The purpose of this quantitative methods study using an online survey
is to make quantitative data available to test Erikson’s theory that what
happens earlier in life affects what occurs later in life as well as the
masculinity theory of STEM of Acker (1990). I intended to establish the
relationship between the number of STEM classes that females take in high
school and postsecondary school and the number of role models females
during these periods of academia have on their career choices and salaries
later in life. Moreover, STEM careers have been traditionally been
considered masculine, and females tend to gravitate towards caring
professions (Acker, 1990; Noddings, 1986). In Research Questions 1 and 2,
the independent variable was career choices and the dependent variables
were the number of STEM classes and same sex role STEM models. In
Research Question 3, the independent variables were the number of STEM
classes and same sex role models and the dependent variable was salaries.
STEM careers tend to have higher salaries than most other fields,
according to the Bureau of Labor Statistics (2009). The objective of my
study was to determine the relationship between the number of STEM
classes taken in high school and postsecondary school and the number of
same-sex STEM role models and their impact on career choices and
salaries, shown in the research questions that I asked in this study.
The purpose was to quantitatively determine the relationship between
the number of STEM classes they took, and the number of role models they
had during high school and college, with their career choices made and the
salaries they receive (Birute, 2009; Farland-Smith, 2009). This career
management is crucial for females to reach their intrinsic and extrinsic goals
(Seibert et al, 2013). The hypothesis was that the more STEM classes
females take and the more same-sex role models females have, the more
likely they are to choose a STEM career and receive a higher salary.
The method was quantitative using an online survey. I analyzed the
data employing an analysis of variance (ANOVA) to look at the
relationship between role models and number of STEM classes taken and
career choices. I employed a linear regression to determine the relationship
between role models and the number of STEM courses taken and salaries.
This study was a relationship study and I did not specifically ask why these
issues exist; the information I gathered possessed clues where these reasons
may be inferred through the specific questions asked on the questionnaire.
The sample was a stratified simple random sample of female alumnae born
after 1980 from four universities on Long Island. Moreover, since the
sample size was drastically reduced, I expanded it slightly by making the
survey also available in the Walden pool of participants.
Research Questions and Hypotheses
The research questions for this study are as follows:
Research Question and Hypothesis 1
Research Question 1: What is the relationship between the number of
STEM courses taken in high school and postsecondary school by females
and their career choices?
Hypothesis One
Ho: The means of the number of STEM classes are the same for the
different career choice categories.
H1: The means of the number of STEM classes not the same for the
different career choice categories.
In this statistical construct using ANOVA, the factor groups were the
career choice categories. The dependent variable was the number of STEM
classes taken, and the independent variable was the career choice categories.
Using a one way ANOVA, I determined if the average numbers of STEM
classes taken were different across factor groups, which were career choice
categories. I aimed to demonstrate that those females who choose STEM
careers tend to take more STEM courses than those who do not choose such
careers. The hypothesis was that there is a positive relationship between the
numbers of STEM classes taken in high school and postsecondary school
and choosing a STEM career. My intention was to retrospectively
demonstrate that the number of STEM courses taken is different by career
choice categories, which are the factors. In other words, I wanted to test if
career choice categories are related to the number of STEM courses taken in
the past. I then compared a set of multiple comparisons to see if some of the
categories were the same statistically.
The original seven categories were defined as follows: the four STEM
groups of science, technology/IT, engineering, and math versus three non-
STEM groups of caring professions, education, and nontechnical. Photonics
and research and development were included in engineering. Caring
professions were healthcare, nursing, medical, and home health aides.
Education was teachers, professors, or anyone who works in a school district
or postsecondary institution. Nontechnical included those professions that
are not in a STEM, caring, or educational profession (including business,
administrative, service, retail, manufacturing, and legal). In the final
analysis, the categories were reduced to five, which were science and math,
IT, engineering, nontechnical careers (soft sciences like business, political
science, education, etc), and caring professions (nursing, social work, health
care, home health aide) as seen in chapters 4 and 5.
Research Question and Hypothesis 2
Research Question 2: What is the relationship between the number of
STEM role models in high school and postsecondary school and female
career choices?
Hypothesis Two
Ho: The number of female STEM role models in high school and
postsecondary school are the same for different career choice categories.
H1: The number of female STEM role models in high school and
postsecondary school are not the same for the different career choice
categories.
Using a one way ANOVA, I determined if the average numbers of
STEM samesex role models are different across factor groups, which are
career choice categories. I aimed to demonstrate that those females who
choose STEM careers tend to have more STEM same-sex role models than
those who do not have such role models. The hypothesis was that there is a
positive relationship between the numbers of STEM classes taken in high
school and postsecondary school and choosing a STEM career
(FarlandSmith, 2009). Under the null hypothesis, using an ANOVA, the
relationships are equal across factor groups. The more same-sex STEM role
models a female has, the more likely she is to choose a STEM career. The
seven original factor groups reduced to five were the same for both Research
Questions 1 and 2 and the intentions are the same, except here I wanted to
see if the career choice categories in retrospect were related to the number of
same-sex STEM role models.
Research Question and Hypothesis 3
Research Question 3: What is the relationship between salaries and
the number of STEM courses taken in high school and postsecondary school
by females and the number of same sex role models?
Hypothesis Three
Ho: Salaries are independent of number of STEM courses in high
school and postsecondary school and/or role models.
H1: Salaries are dependent on the number of STEM courses in high
school and postsecondary school and/or role models.
The number of STEM courses and the number of same sex role
models were the independent or predictor variables, and salaries was the
dependent or outcome variable. This was a multiple regression. In this
research question, I attempted to establish a relationship between the
independent and dependent variables (Aczel & Sounderpandian, 2009; Field,
2013). The hypothesis was that the number of STEM classes taken and the
number of same-sex STEM role models have a positive relationship with a
higher salary since if these conditions exist, it is more likely females will
choose STEM careers that tend to have higher salaries.
Theoretical Framework
In this study, the theoretical framework was based on the theory of
social development of Erikson (1980) and Acker’s (1990) masculinity
theory. According to Erikson’s theory of social development, events that
happen in childhood, such as exposure to certain areas of interest like the
sciences, may affect the career choices made later in life (Erikson, 1971,
1980, 1997). Educational development in the STEM fields that females
obtain in their high school and college education may positively influence
their career choices into higher paying fields. Furthermore, the play
activities females partake in as children may impact their later interests in
science as a career because engaging in science activities as youth increases
interest in science, which may increase the chance of choosing a STEM
career (Erikson, 1980; Noddings, 1986). When girls have an interest in
science at a young age, they may be more likely to take the STEM courses
and choose a STEM career (Farland-Smith, 2009; Klawe, 2013). In this
study, the emphasis was on the relationship between the number of these
STEM classes that females take as well as the number of role models they
have starting in high school and continuing at the postsecondary level. These
early events are affected by this social development theory. Additionally,
the analytical background the females gained in their social and educational
development could be a possible indicator of whether or not they chose to
take more than 3 years of STEM classes (Erikson, 1980, 1997).
In an example of how females were segregated from STEM
backgrounds in their early educational development, Irby and Brown (2011)
asserted that white middle class children at a school in Great Britain were
segregated by gender, and males were encouraged in competitive play, math,
science, technology, and tasks of leadership, assertiveness, and power.
Females were discouraged from these STEM subjects and tasks. Irby and
Brown postulated that this segregation could possibly contribute to lower
paying careers and caring roles of females in the generativity stage as well
as not taking many STEM classes in college during the intimacy stage of
Erikson. This theory was also postulated by Gilligan,(1986) and Noddings,
(1986). This analysis by Irby and Brown as well as my study contributes to
understanding why females may have different social or professional
experiences later in life. These early life experiences may impact their career
choices during the both the sixth stage, known as the intimacy stage, and
seventh stage, known as the generativity stage of Erikson’s theory of the
eight life stages of development (Aldwin, 2009). The sixth stage is where
people are young adults who establish committed relationships to begin
families during the ages of 19 to 40, as well as attend college and plan and
begin their careers. The generativity or seventh stage consists primarily of
parenthood, work, and family, where careers are established (Erikson, 1971;
Gilligan, 1986). These two stages overlap.
Instead of strictly using a sample between the ages of 19 to 40, I
concentrated on a sample of those in their early 30s, born after 1980, which
are in the sixth or early part of the seventh stage of Erikson’s (1980) life
stages. At this time, both male and female adults learn generatively, seeing
the world through a more global perspective. However, females use this
global perspective to enhance the ethic of care through altruistic roles
(Erikson, 1971; Gilligan, 1986; Noddings, 1986). The concept of altruism,
according to Erikson (1971), was derived from the early stages of
development based on societal norms when girls are expected to play and
take care of their dolls like mommies (as cited in Gilligan, 1986; as cited in
Noddings, 1986). As adults, according to Erikson and Noddings (1986),
females usually became the primary caregiver of the family, including the
extended family. These factors may also impact whether or not females will
take the necessary number of STEM classes needed to enter such a career
and make such a career choice. These early events may also contribute to the
lack of role models females have to emulate in these STEM fields. Irby and
Brown (2011) also felt that those females with a strong analytical
background that began since childhood are more likely to take STEM
classes and choose math and science careers. Acker (1990), Erikson (1980)
and Noddings (1986; 1995) are in accord with Irby and Brown.
Acker (1990) also developed a theory where she postulated that
females working in jobs that are traditionally male (Royal, 2007) had an
uncomfortable self-image and self-esteem performing functions that are
opposite to what they naturally do to perform. According to Acker, the
natural job functions for females were to gravitate towards more caring
professions (as cited in Noddings, 1986). These females are expected to
exhibit the same behavior a male would in the same role (Acker, 1990).
Females who do attain
STEM degrees, particularly in computer science and engineering, experience
the glass ceiling, making it more difficult to become leaders or managers
(Dugan et al, 2013; Kellerman & Rhode, 2007; Noddings, 1986). For this
reason, it is important that females are more encouraged to enter STEM
fields early in their education and academic careers by participating in more
scientific activities in childhood and taking more classes in high school and
college. There is also a reference to these theories in Chapter 2.
For the purpose of my study, the levels of STEM classes begin in high
school with ninth grade general science, algebra, and computer classes
through 12th grade physics and precalculus, on to postsecondary from
freshman precalculus, to undergraduate calculus, differential equations, to
masters level math. The same levels represent the sciences from basic
freshman biology to higher level undergraduate anatomy and physiology,
chemistry, meteorology, geology, and all other branches of sciences taken
from the undergraduate to the masters level or doctoral level.
Nature of the Study
In this study, I employed an online survey for data collection, which
was a crosssectional design (Campbell & Stanley, 1963; Harris, &
Finkelstein, 2006). I chose a quantitative study because this is a relationship
and correlation study that is more effective quantitatively than qualitatively
(Creswell, 2014). This method also increases the validity and reliability of
the results. This method inherently makes it difficult to control extraneous
variables (Case, 2007; Shao, 2002). In the survey, I employed a valid and
reliable 5-point Likert scale (Becker, 1986; Reynolds, 2007). The Likert
scale is an ordinal, permitting ranking that measures attitudes (Nachmias &
Nachmias, 2008; Shao, 2003). The sampling frame was originally 487
female respondents from four universities from their alumni lists, born after
1980 (Kalton, 1983). However, the low response rate resulted in a small
dataset of 48 respondents. There was a pilot study for testing the survey
before the research began (Nachmias & Nachmias, 2008; Teijlingen &
Hundley, 2001; Yin, 2003).Since the sample size was reduced, I expanded it
slightly by making the survey also available in the Walden pool of
participants.
The focus in this study was females who were born after 1980 and
who are members of the Millennial generation. The sample was a stratified
random sample drawn from four chosen Long Island universities’ alumni
associations. These four universities and the female alumnae, along with the
small Walden pool of participants were the strata of the population, making
the sample stratified (Kalton, 1983; Nachmias & Nachmias, 2008; Rea &
Parker, 2014; Shao, 2002; Yin, 2003). By choosing this population, I
evaluated this relationship based on females who are currently in their 30s to
determine why the gender gap in STEM careers may persist based on the
relationship between courses taken and career choices.
The method was an online survey instrument, which was a cross-
sectional design, according to Campbell and Stanley (1963). Survey research
works well with either random or simple random samples (Kalton, 1983),
like mine. This method is the most appropriate for this study because this
was a relationship study making inferences between a predictor and an
outcome variable (Campbell & Stanley, 1963; Nachmias & Nachmias, 2008;
Survey 2005). According to Babbie (2006) and Nachmias and Nachmias
(2008), the quantitative cross-sectional research design is most appropriate
for survey research. Survey research is normally conducted through mail
surveys, personal interviews, or telephone interviews. However, in person
and phone interviews were too qualitative in nature for my study. In my
case, the survey was conducted online.
According to Shao (2002), closed-ended questions are used in the
quantitative approach. My survey was made up of closed ended questions
using a 5-point Likert scale, which is quantitative (Shao, 2002). An
ANOVA or a regression was developed to analyze the relationship among
the variables (Field, 2013). An ANOVA is a way of breaking down the total
variability into smaller categories or components and assessing if the
variability due to a specific source is statistically significantly higher than
the random variability (Green & Salkind, 2011).
The ANOVA compares the differences in the means of salaries and
career choices across the categories. For a one-way ANOVA, each
individual or case must have scores on two variables, factors, and
dependent variables, which divides individuals into groups across
categories (Field, 2013, p. 183). The ratio of these variances is known as the
Fratio (Field, 2013). The regression can analyze the strength of the
relationship between the variables (Field, 2013). The ANOVA may help
quantify career choices, especially since I conducted a cross-sectional
survey (Field, 2013; Nachmias & Nachmias, 2008).
In quantifying these theories by Erikson (1980) and Acker (1990), the
number of STEM classes taken by females is already quantified because it is
the number of classes taken at the high school and postsecondary levels to
the graduate level, which are all quantified. However, the challenge was
quantifying the concept of career choices. To quantify, I created groups of
career choices where the measure was the number of STEM courses taken
by each participant. Subsequently, I determined with an ANOVA which
career choices had the highest average number of STEM courses. The
averages were ordered. Then a post hoc indicated which differences are
significant (Field, 2013).
The hypothesis was that the higher the number of STEM classes that
females take in high school and postsecondary school, the more likely they
will choose STEM careers and increase their salaries. Salaries are already
quantified and were used in numerical categories (Babbie, 2006). The
number of these STEM classes that females take impacts whether or not
females choose a STEM career and influences whether they were
encouraged to take such classes or enter such career, if they had role models,
if they believed they have the abilities for math and science, or if they
experienced societal discrimination. Moreover, another factor is whether or
not females were encouraged in the sciences as children. Furthermore, the
communications or behavior or how females view their careers could also
impact on how many math and science classes they take and the career
choices they make. The challenge was quantifying career choices. This can
be accomplished through an ANOVA (Field, 2013).The regression was
employed to determine how close the relationship is between the predictor
variables to the one outcome variable.
The independent variable in Research Questions 1 and 2 was the
career choices and the dependent variables were the number of STEM
classes and the number of female STEM role models because the literature
suggests that the lack of female STEM role models may correlate with
females avoiding taking STEM classes or choosing such careers (Farland-
Smith, 2009). In Research Question 3, the number of STEM classes and the
number of same sex role STEM models are the independent or predictor
variables, with salary as a dependent variable.
Due to the nature of the variables in this study, the instrument is best
used with interval scales (Field, 2013; Nachmias & Nachmias, 2008). The
survey used a valid and reliable 5-point Likert scale using both interval and
ordinal scales (Becker, 1986; Reynolds, 2007; Shao, 2002). The Likert scale
is ordinal and interval, permits ranking, and uses a continuum (Nachmias &
Nachmias, 2008; Shao, 2002) with a specific research design. There are four
steps to consider with the Likert scale. First I compiled the scale items. Then
I administered the scale and questions to the chosen sample for the survey.
Then I computed the value of the scale with the first response as 1, the
second as 2, the third as 3, the fourth as 4, and the last as 5 and then
summing up the values. From there, I determined the discriminate power by
taking the highest and lowest values and determining the differences
between them (Trochim, 2006a). Finally, I selected the highest power
discriminates selected and tested the reliability of the scale, as explained by
Nachmias and Nachmias (2008).
There was a pilot study testing the survey before the research began
(Teijlingen & Hundley, 2001). The purpose was to test the instrument to
increase reliability because the questions may have needed to be modified in
order to answer my specific research question (Becker, 1986; Reynolds,
2007; Shao, 2002). After the research was conducted, there was a posttest
(Teijlingen & Hundley, 2001; Yin, 2003). The comparison with males who
took science, math, and technology classes was obtained from the vast
existing data, and the comparison was with females who took less than 3
years of STEM classes since high school and those who took more than 3
years. According to Campbell and Stanley (1963), a pilot survey improves
the survey instrument.
For the quantifying of the career choices, there was a group of career
choices where the measure was the number of STEM courses taken by
females in the sample. Subsequently, I determined using an ANOVA which
career choices have the highest average number of STEM courses, to
quantify career choices. The averages were ordered and a post hoc indicated
which differences are significant (Field, 2013; Nachmias & Nachmias,
2008). For the salaries, a regression was performed to determine if the
number of STEM classes a female student takes and the number of female
role models she has may be valid predictors of her salary. The question was
the same for the number of female role models. Subsequently, I employed
salary groups as used in marketing research, and these groups were the
factor (Belch & Belch, 2004). The number of STEM classes was the
predictor of career choices, and then I conducted a one-way ANOVA. I also
quantified the number of role models related to the career choices. The
theory that my theoretical framework is based on is Erikson’s (1980)
concept that what happens early in life impacts decisions later in life.
Furthermore, Correll (2004) believed that culture about the masculinity of
math may have discouraged females from taking these classes.
To investigate whether math and science classes influence female
career choices, an internet survey tool was used. Since I designed my own
survey questions, the pilot study served to test the reliability of my
questionnaire (Shao, 2002; Yin, 2003). I discussed my rationale for using
this designed questionnaire by discussing how the instrument was used
before and for what population it was used. I discussed what the tool
measured and how it applied to this study.
The Significance of the Study
This is a valuable study that can bring about social change. If females
are encouraged to and do take more STEM classes in their social
development of education, they can obtain the training to be able to make
STEM career choices (Wrigley, 2002). This study is significant because it
may increase the understanding as to why there is a gender gap in STEM
fields and how to close this gap. Addressing such issues as the thought that
females are not as good in math, or lacking role models in STEM fields may
be addressed, thereby helping females to increase their access to these higher
paying careers. Females can see how important science and math are early in
life and how parents should encourage their daughters to be interested in
math and science as children. By making quantitative data on the
relationship between the number of STEM classes females take, the number
of female role models, and the impact on career choices and salaries
available, this information might help females to better manage their course
selections to be competitive in their career choices within the STEM field.
These data might also help guidance counselors and deans to aid females on
counseling on how to better manage STEM careers, both academically and
in the workplace in this broad science of management.
Significance to Management
The field of management is a broad social science that has many
functions, including planning, organizing, staffing, controlling, budgeting,
evaluating, and implementing. The functions that apply in this study have to
do with setting goals (Drucker, 1954), planning careers, organizing these
goals, evaluating, and implementing them. When females manage their
academic and career goals, they are using these functions. Some clear
principles of management applied here are teamwork, team learning, long
and short term career planning, and leadership (Kellerman & Rhode, 2007;
Senge, 2006). When females can access STEM careers and close the gender
gap, this is a continuous improvement in their career, similar to the concept
of total quality management that Deming (1960) discussed.
Team work, team learning, and collaboration are more important to
females than individual self interest (Senge, 2006). According to Manning
(2012), a team’s legacy is more important than that of a specific individual.
Moreover, Jiang (2010) believed that collaboration fostered community and
teamwork. Klawe (2013) discovered these ideas in her study at Harvey-
Mudd College for Computer Science when she studied a sample of female
students and surveyed them to find out why they were not taking computer
science courses. What she discovered was that females saw computer
science as boring, difficult, and designed specifically for geeks, and not team
or community oriented. It seems that females prefer community, team work,
and collaboration over individualism (Kellerman & Rhode, 2007; Noddings,
1986; Senge, 2006). Learning as a team or a community makes females feel
as if they belong, and this motivates them towards STEM skills, classes, and
majors (Klawe, 2013; Senge, 2006). Therefore, this aspect of teamwork and
team learning was very applicable to the field of management and to my
study. Motivation is also crucial to management, especially human
resources, and this concept of teamwork helped to motivate these females to
take computer science, according to Klawe. Moreover, self-efficacy is
defined as believing in one’s self in obtaining an academic or career goal
with one’s own initiative and self determination (Bandura, 2003; Raelin et
al, 2014).
Career planning and career management are also very important and
significant to the field of management, and I touch on this concept
frequently. I investigated the relationship between the number of STEM
classes females take in high school and postsecondary education and the
number of female role models and how these two variables impact on career
choices that females make and their salaries. There are several reasons why
career planning is important to management. Planning is the first and most
important management function, which involves MBO, the setting of long
and short term goals (Drucker, 1954). Management also involves setting
intrinsic and extrinsic goals (Seibert, et al, 2013). Furthermore, management
involves setting intrinsic and extrinsic goals (Seibert et al, 2013). Career
management also involves evaluating one’s career goals beginning with the
classes one must take. For my study, I examined the STEM classes that
females take. It is evident that the more classes they take in the STEM field,
the more experience, knowledge, and training they receive in these careers,
making it more likely that such a female would choose a STEM career,
which is one of my hypotheses. I also examined the role models that a
female has in the STEM field. The more role models that a female has, the
more likely she may choose a STEM career (Farland-Smith, 2009), which is
also one of my hypotheses. Moreover, according to the Bureau of Labor
Statistics (2009), STEM careers offer higher salaries than other careers, thus
if a female chooses such a career, I hypothesized that she may earn a higher
salary than if she entered a caring profession (Noddings, 1986). By making
this quantitative data available, this information may help females to better
manage their course selections to be competitive in their career choices
within the STEM field. These data can also help guidance counselors and
deans to aid females on counseling on how to better manage STEM careers,
both academically and in the workplace as the scientific and management
leaders of tomorrow.
Leadership and management were applied in this study because many
STEM positions also have leadership roles or may lead to leadership
positions. Kellerman and Rhode (2007) saw any gender gap in careers as
negatively impacting female access to leadership and management positions
as well as in accessing STEM positions. As a result, females enter both
leadership and STEM careers at lower rates than their male counterparts
because they may have fewer resources to take the necessary classes, lower
self-concept of their abilities, and may be discouraged from these classes
early in their childhood or academic careers. Therefore, females have been
discouraged from choosing these careers, which may possibly negatively
impact their salaries, as well as access to leadership positions in these fields
(Kellerman & Rhode, 2007). It was evident that this study had many
applications making it significant to the field of management as well as
significant to my profession as a business and management instructor.
Significance to Profession
As an adjunct instructor of business, marketing, and management, I
used many of the principles of management previously mentioned,
especially career planning, in the classroom. In each of my undergraduate
management and marketing classes, I conduct a unit on careers in that
subject, and I also discuss issues of gender bias, gender gaps, and
discrimination in the work place with my students. This study may be very
helpful as a teaching tool that can be used to supplement some of the
material in my management classes. Moreover, I encourage my students to
take as many STEM classes in college as well as business classes to broaden
their knowledge and increase their chances of being able to access one of
these high-paying careers.
Furthermore, in my undergraduate management classes, I teach the
concepts of many aspects of management focused in this study. These
include the concepts of leadership and management (Kellerman & Rhode,
2007), team work, team learning, and collaboration (Argyris, 2003;
Manning, 2012; Senge, 2006), and career planning as well as MBO
(Drucker, 1954). In my management classes, I review all of these mentioned
theorists, their theories, and practical applications. More importantly, when
discussing career planning, I cover gender gaps in the work place as well as
gender discrimination and gender segregation in the work place. Also in my
management, marketing, and business classes, I cover the reasons for gender
gaps in careers and wages, as well as gender discrimination and segregation
in business.
Definitions and Terms
STEM classes, role models, career choices, and salaries are the
variables used, and all but salaries are defined. There will be more detail in
Chapter 3.
Career choices: The chosen career fields that females chose who have
taken more STEM classes and the salaries they receive as a result of these
choices. The STEM classes females took became the dependent variable and
the career choice was the independent variable or the outcome variable
based on the classes taken for Research Question 1 and the same-sex STEM
role models the females had in the past (Field, 2013). The salary was the
dependent variable, and the number of STEM classes and female role
models were the independent variables in Research Question 3 as such a
career choice may also correlate with an increased salary level. Career
choice categories were grouped by science, technology/IT, engineering,
math, education, caring professions, or nontechnical fields in the first two
research questions.
Gender segregation: The separation of the sexes since their childhood
based on gender. This is where children play with same-sex peers and with
sex typed playthings appropriate for their gender (Kellerman & Rhode,
2007). Even today, 55% of girls and only 2% of boys play with dolls.
Conversely, 41% of boys play and only 4% of girls play with toy vehicles
(Townsend, 2013). This kind of gender difference can affect socialization
and impacts decision-making later on in life, leading females towards caring
professions and leadership styles, and males into more competitive
professions and leadership styles (Dugan et al, 2013; Kellerman & Rhode,
2007). Socialization is the personal and social interaction of males and
females based on gender.
Role models: Defined as a female who works in the STEM field who
has helped to inspire or encourage the female respondents to take additional
STEM classes in high school and college and to choose a STEM career.
Role models can be a mother, aunt, cousin, friend, grandmother, teacher,
professor, colleague, or employer. Role models are anyone who either
directly or indirectly influenced career choices or education majors through
either admiration or emulation.
Science, technology, engineering, and math (STEM) classes: These
include higher level math from algebra to calculus and differential
equations. Science includes hard sciences like biology, anatomy and
physiology, archeology, chemistry, physics, astronomy, aeronautics,
astronautics, or geology, life sciences, archeology, and astrophysics. It also
includes certain soft or social sciences such as psychology and sociology,
but not political science, economics, or business. However, for this study,
the soft sciences were put under the category of nontechnical. Technology is
software, hardware, IT, cloud computing, software engineering, and
anything related to computers. Engineering means design and research and
development, mechanical, electrical, aeronautical, astronautical, aerospace,
and photonics. Math includes general math, finite math, algebra, geometry,
trigonometry, precalculus, calculus, linear matrix algebra, statistics,
econometrics, and differential equations
Scope of Study
This study has a scope that is limited to a sample of approximately 48
female alumni from four different Long Island universities, or the number of
universities who cooperated with my study, randomly chosen by the alumni
association, who were born after 1980. The only qualifier was that they be
female and born after 1980. The scope is limited to this geographical area,
and there might be some issues with generalizing to the general population.
The scope is broad enough, however, to include all females in these
universities regardless if they have taken STEM classes or not or had role
models or not, because I wanted to investigate the relationship between the
number of STEM classes taken and the number of role models from none to
an infinite maximum in order to determine the impact on career choices and
salaries.
Positive Social Change Implications
The social change implications may be to reduce the gender gap in
STEM careers and help females have access to the same high paying STEM
careers as males. This can be accomplished through improved training in
STEM (He & Freeman, 2010) and increase encouragement into these fields,
which may indirectly increase the role models (Farland-Smith, 2009). By
increasing these opportunities for females, more females may choose these
careers, and there may be an increase in role models to encourage more
females to choose STEM careers.
The social change implication can be how society may help to create
more opportunities for females to enter STEM careers through improved
training and encouraging science and math interest in early childhood and
early on in their academic careers. Moreover, these new opportunities for
females may result in the availability of more role models to encourage them
to make STEM career choices (Farland-Smith, 2009; Milgram, 2011;
Noddings, 1986). It is hoped that with more females in STEM careers,
attitudes may also become more favorable towards the abilities of females in
these careers, bringing about social change.
Assumptions, Limitations, Delimitations
Assumptions
Whether or not a female was encouraged to take more STEM classes
and careers in school or had role models could influence the salary she
receives and career choices she makes, which is an assumption (Correll,
2004; Wrigley, 2002). For Research Question 3, the independent variables
were the number of STEM courses females took and the number of female
role models they had, and the dependent variable was salaries (Correll,
2004; Kenney et al, 2012). In Research Questions 1 and 2, the variable
called career choices was the independent variable with the number of
STEM classes and role models as the dependent variables. Career choice can
be quantified because I allocated responses to a group of career choices
where the measure was the number of STEM classes taken. Then an
ANOVA was conducted to quantity and determine which career choice
groups need the highest average of STEM classes in retrospect. Then, I
ordered the averages, and conducted a post hoc to determine which
differences are significant. The number of role models and the number of
STEM classes were the independent variables, and salary was the dependent
variable in Research Question 3. There were also assumptions that females
may leave the workforce for any reason. Here it was possible to use an
ANOVA where leaving the workforce can be an auxiliary variable known as
length of time out of workforce (Field, 2013; Morrow, 2013). I also needed
to reduce threats to internal validity such as maturation and morbidity by
keeping the survey reasonably short, less than 30 minutes to fill out (Field,
2013; Nachmias & Nachmias,
2008; Shao, 2002).
It was also assumed that when contacting the alumni associations of
the four sampled school strata, there would be cooperation between the
alumni association and myself in disseminating the surveys to the students
as randomly as possible. The associations were asked to contact the students
due to confidentiality through email or a newsletter, which were conducted
randomly or using a systematic interval where the starting point is random.
In both of these systems, each student had an equal chance of participating
(Shao, 2002); therefore, I let the association determine what was easiest for
them to ensure cooperation.
Limitations
Some limitations included that the sample was limited to only four
Long Island universities; it was difficult to obtain a cross section of the total
population, based on a localized area, with a limited geographic scope. This
could make it difficult to generalize to the entire population affecting
validity (Nachmias & Nachmias, 2008). Furthermore, this was a correlation
study, which means that causation cannot be determined. I cannot make
claims that the lack of STEM classes that females take correlates with them
to choose careers outside of the STEM fields. I could only hypothesize that
the more courses they take and the more role models they have should have
a positive relationship to them choosing STEM careers and receiving higher
salaries. Other limitations that could not be controlled are the financial and
mobility constraints of the researcher. For these reasons, the study was
conducted online using an online survey instrument. I needed to make sure
the questions were objective and as valid as possible with a Cronbach alpha
of.7 or .8. Since it was lower, I needed to make some adjustments to the
questions to reduce bias.
Delimitations
Delimitations are the factors that I as the researcher have chosen,
which are the boundaries I have set for this study. The first boundary I have
set was that I only considered females born after 1980, living in Long Island
who was alumnae of the four universities chosen for this study. The sample
was randomly chosen by the alumni associations of the four universities I
have chosen for this study along with the few from the Walden pool to
counteract the reduction in the sample from 487 to 48. The instrumentation
was an online survey, and the reason this study was online was to control
cost and also because I have difficulty with mobility. Moreover, online
surveys are easier to administer, more global, cost effective, and have higher
response rates than postal mail surveys (Patton, 2009; Shao, 2002).
Practical Implications
The results of the study indicated that taking more than 3 years of
STEM classes in high school and postsecondary school and having female
STEM role models correlate positively with career choices, although, the
correlation with role models was weaker than that of STEM classes and
career choices. This should help females obtain the training necessary to
impact their decisions to pursue these career choices. A major benefit for
females could be higher pay as a result of being able to make career choices
in the STEM fields. This is a practical benefit because females need to pay
bills, earn a living, and save for retirement. If females are given more
opportunities to take STEM classes, then females would able to impact their
decisions to pursue career choices in these higher paying fields and reduce
the gender gap in STEM fields as postulated and agreed by Carrell, Page,
and West (2010), Farland-Smith (2009), Gilligan (1988),
Noddings (1986) and Sharp et al (2008).
Summary of Chapter 1
As an overview of Chapter 1, I gave an introduction of the study and
provided some background in order to give the reader some rationale for the
study. In this chapter, was the problem statement, and the need or reason this
study is important. I have also covered the goals and objectives of the study.
Moreover, the research questions, variables, and the hypotheses were
introduced in order to transition into the literature review. There was also an
introduction to the methodology in order to help the reader better understand
the data collection and analysis methods used in this study. An objective is
to give an introduction to what will be further explained in Chapter 3.
The theoretical concept was based on the developmental theory of
Erikson, particularly Stages 6 and 7. Stage 6 is the intimacy stage where
people attend college and begin their careers as well as steady relationships.
Stage 7 is the generativity stage when people reach middle age and create
commitments to family and careers. Most people peak in their careers during
this stage. Also, the theory was based on what happens when females are
children and what affects their STEM classes and career choices in Stages 6
and 7. For example, if they are not encouraged in math and science as
children, they are less likely to pursue STEM majors or careers later on in
life.
In this chapter, I covered the significance this study has to
management and my career as a business, management, and marketing
instructor. The main applications used in management and in my career that
were focused upon in this study were leadership, team work and team
learning, management, leadership, management by objective, and career
planning as well as the basic management functions. Lastly, the scope, the
limitations, assumptions, delimitations, implications, and social change
impacts were also covered, leading into Chapter 2, the literature review.
Chapter 2 is the literature review that sets the groundwork for the
theoretical framework employed in this study. Furthermore, it is the
background analyzing the literature that led to the gap in the literature that
this study expects to fill. In chapter 3, I outline the data collection and
analysis quantitative research methods employed in this study. Chapter 4
offers the findings and the results of the analyses, indicating to what extent
the hypotheses were supported or not. Chapter 5 begins with the summary of
study’s results with an analysis of the study’s implication of social change
and on scholar-practitioners.
Chapter 2: Literature Review
Introduction
One way to promote positive social change is to bring to the forefront
a persistent gender gap in the fields of STEM. A goal of my study is to
educate the academic community on how to help reduce barriers that have
traditionally kept females from entering these fields. For this reason, Chapter
2 laid out the groundwork for this study beginning with a discussion about
how there are still fewer females entering STEM professions as opposed to
males even today. Chapter 2 provided a review of scholarly literature to lay
the groundwork for this study. The problem was that a persistent gap exists
between the number of females and males accessing high-paying STEM
related careers as well as high paying management career (Hensvik, 2014).
Despite antidiscrimination measures, laws, and social progress for females,
they still are not entering STEM careers at the same rate as their male
counterparts (Carell, Page, & West, 2010).
Females are still not taking as many STEM classes or majoring in
these fields in high school and college at the same rate as males to prepare
for employment in these fields, according to Carell et al (2004), Moakler and
Kim (2014), and Milgram (2011). According to Brown et al (2011), the
proponents of STEM education believed that by increasing math and science
requirements in schools, schools partnering with local technical businesses,
along with incorporating technology and engineering concepts into
curricula, students will perform better and be better prepared for advanced
education or jobs in STEM fields, which is often referred to as the STEM
pipeline. Bystydzienski, Eisenhart, and Bruning (2015) postulated that high
school is not too late to augment an interest in STEM for females,
particularly in engineering. This STEM pipeline concept would encourage
females to enter more STEM careers if they took more of these classes.
These remedies would help increase the amount of females entering STEM
fields.
Since these remedies have not been fully implemented, females are
not entering STEM fields at the same rate as their male counterparts.
However, they are entering fields like medicine and law at increased rates.
According to Friedman (2008), females tended to enter medicine and law
more than they enter STEM careers. Although males still enter medicine and
law more than females, females are making strides in these fields at a faster
rate than in the STEM fields (Friedman, 2008). In fact, Bystydzienski et al,
(2015) found that although the participants began high school with little or
no knowledge of engineering, it was easy to develop their interest, which led
them to seriously consider engineering as a college major and future career,
but only 18% of the female participants resulted in choosing an engineering
career in this longitudinal, after-school intervention study.
Furthermore, female STEM role models are crucial to inspiring
females to enter science related careers (Bystydzienski et al., 2015).
Milgram (2011) stated that role models that are similar to young female
students play a key role in young women’s decisions whether to go into the
science, math, or technology field (Acker, 1990; Drury, Siy, & Cheryan,
2011). Role models must be females. They can be relatives such as aunts,
cousins, mothers, or they can be a teacher, professor, or employer who
inspired a young woman to enter a STEM related career. Even President
Obama, in collaboration with the Girl Scouts and NASA, have partnered to
encourage young females from age 7 to 18 to participate in science fairs,
where they can come up with their own realistic science projects through
projects and annual science fairs at the Whitehouse and meet same sex role
models in these fields (Byron & Nye, 2014), making science fun (Drury et
al., 2011; Farland-Smith, 2009; Milgram, 2011). Some of these projects
included rocketry, robotics, electric cars, photonics, and medical
applications. With projects like this, the hope is to increase STEM
employment opportunities for females, which would also increase the
number of same sex role models available to young women looking to enter
STEM fields.
It has been more difficult for females to find role models who may
encourage them to take STEM classes and enter these science fields. That is
why it was crucial to study if there is a relationship between the number of
STEM classes taken by females and the number of same sex role models and
their career choices along with salaries. I inquired the reasons this gap
continues, based on this relationship, which is the problem of interest in my
study. This problem brings the discussion to the research questions that I
asked in this quantitative study.
Through the research questions, I examined the relationship between
the STEM classes that the sampled females took in high school and
postsecondary education, the role models, if any, they had and how these
items relate to their career choices, and their salaries. From the literature
review, I have developed the independent variables, which are the number of
STEM classes taken and female STEM role models in high school and
postsecondary school as well as middle school in Research Question
3(Bystydzienski et al., 2015; Drury et al., 2011; Farland-Smith, 2009). The
dependent variable was salaries for Research Question 3. For Research
Questions 1 and 2, the independent variable was career choice categories,
and the dependent variables were the number of STEM classes and role
models, using an ANOVA. In Research Question 3, I employed a regression.
I used an online survey using a 5-point Likert scale, which is a
standardized, valid, and reliable ordinal and interval scale used universally
(Comley & Beaumont, 2011; Nachmias & Nachmias, 2008; Shao, 2002).
For the data analysis, since this was a relationship study, I conducted an
ANOVA to quantify career choices; I also conducted an ANOVA since the
survey is cross-sectional (Field, 2013), and a linear regression was used for
the other variables.
Literature Search Strategy
In this literature review, I demonstrate and analyze the relationships of
why many females have avoided STEM careers, and I discuss the external
environment of these females, how they communicate, and how this gender
gap became established (He & Freeman, 2010; Milgram, 2011), based on
prior literature. This led up to the rationale for my study based on what has
already been researched in the body of literature and where the gaps are that
my study can fill.
When I was searching the databases, I used the following key words
and phrases in the subject line in order to search the literature: STEM
courses and females, STEM career choices, female STEM employment,
relationship of STEM classes to career choices, female STEM role models,
and career choices. I searched the Walden databases including Business
Primer Complete, Thoreau, Proquest, and Ebsco. In the search for more
current articles, I used qualifiers such as full text, peer-reviewed, and from
the years of 2008 to 2014. Some scholarly journals I consulted included
Gender and Society, Public Relations Quarterly, Research in Human
Development, Academic and Educational Leadership Journal, Mid-
American Journal of Business, Career Development Quarterly,
Harvard Business Review, Science Education, International Journal of
Business Management, and others. If I knew the exact title or digital object
identifier (DOI), I would use the find exact article feature in the Walden
library. Moreover, I searched multiple databases including the business
database called ABIInform and other business databases as well as
interdisciplinary databases. This helped to ensure that the majority of the
peer-reviewed articles were from current literature in business, education,
human resources, and management because this topic is quite dynamic
where progress is bringing about continuous social change. There were a
few ancillary articles used from journals such as Atlantic Monthly, Photonics
Spectra, and Business Week, which although these journals were not peer-
reviewed, they offered some important points in the field of business and
management and demonstrated some important current trends in the gender
gap in STEM fields in employment and salaries. There were also a few
videos used that made important current comments on the gender gaps in
STEM and helped to enhance this literature review. However, the majority
are peer-reviewed articles from scholarly journals. In cases where there was
little current research and few if any dissertations and/or conference
proceedings within the last 5 years, I used the next most recent studies or
identified the gaps my study can fill.
This literature that I analyzed in this review was the academic base for
my research, examining what prior research has been conducted, and what
gaps my research can fill in the body of knowledge. I have chosen sources
that offer some prior studies on the research questions and can help me find
the gaps where my study can add to the body of knowledge. The purpose of
this literature review is to investigate and to critically analyze the literature
and determine the gaps to see where my study fits in the body of literature
(Randolph, 2009).
I identified several key themes in the literature. First, the literature
comprehensively covers some of the causes and effects, correlations, and
frameworks of research that have been conducted before on the gender gap
in STEM careers and the relationship between STEM classes taken and
career choices and salaries. Secondly, I viewed the literature for the reasons
for the gender gap (Moakely & Kim, 2014), for females taking fewer STEM
classes than males and having few same sex role models.
Thirdly, the articles in this review offered a basis for the survey tool
as the research method that I used in this dissertation (Shao, 2002), as well
as a basic literature review on some of the research already conduced on the
relationship between STEM classes taken by females and their career
choices and their salaries. Furthermore, I examined any survey tools that
were similar to the online survey tool used for my study and evaluated the
strength and weaknesses of the on line survey tool and evaluated articles on
the issue of females and STEM careers to help answer the research question
by offering a background on reasons for the gender gap in STEM careers
(Kenney et al., 2012; Moakley & Kim, 2014). Knowledge obtained about
quantitative data collection methods like the online survey and the 5-point
Likert scale came from Creswell (2014),
Kaczmarek et al. (2012), and Shao (2002).
Theoretical Foundation or Conceptual Framework
In this study, the theoretical framework was based on the theory of
social development of Erikson (1980) and the masculinity theory of Acker
(1990). According to the theory of social development, events that occurred
earlier in life affect the choices made later in life (Erikson, 1971, 1980,
1997). The educational development in the STEM fields that females obtain
in their high school and college education may positively influence their
career choices into these higher paying fields. Furthermore, the play
activities females partake in as children may impact their later interests in
science as a career, on their later interests in science as a career because
engaging in science activities as youth increases an interest in science that
may increase the chance of choosing a STEM career (Erikson, 1980;
Noddings, 1986). When girls have an interest in science at a young age, they
may be more likely to take the STEM courses and choose a STEM career, as
evident from studies conducted by Bystydzienski et al., (2015), Drury et al.
(2011), Farland-Smith (2009), and Klawe (2013), which demonstrated
evidence of Erikson’s theory in these past studies. In this study, the
emphasis is on the relationship between the number of these STEM classes
that females take as well as the number of role models they have starting in
high school and continuing at the postsecondary level and their career
choices and salaries. According to Moakley and Kim (2014), females avoid
STEM because of the lack of female role models in their early education
(Moakley & Kim, 2014). These early events are affected by this social
development theory. This developmental theory applies because I
hypothesized the classes and role models females have in their early years
have an impact on the career choices they make and the salaries they
potentially earn. The theory of the analytical background the females gained
in their social and educational development could be a possible indicator of
whether or not they chose to take more than 3 years of STEM classes
(Erikson, 1980; 1997).
Acker (1990) also developed a theory where she postulated that
females working in jobs that are traditionally male they had an
uncomfortable self-image and self-esteem performing functions that are
opposite to what they naturally do to perform. Royal (2007) disagreed with
this concept and felt that females should work in these jobs to build their self
image. According to Acker, the natural job functions for females were to
gravitate towards more caring professions as discussed also by Noddings
(1986) and not technical fields like He & Freeman (2010) postulated. These
females in these nontraditional jobs are expected to exhibit the same
behavior a male would in the same role (Acker, 1990). Females who do
attain STEM degrees, particularly in computer science and engineering,
experience the glass ceiling, making it more difficult to become leaders or
managers (Dugan et al, 2013; Kellerman & Rhode, 2007; Noddings, 1986).
For this reason, it is important that females are more encouraged to enter
STEM fields early in their education and academic careers, by participating
in more scientific activities in childhood and taking more classes in high
school and college. This rationalizes investigating the kind of relationship
between the number of STEM classes and role models with career choices
and salaries. Only He and Freeman (2010) and Alshare and Miller (2009)
postulated that because of this expectation in behavior, many females
avoided STEM courses in school and careers because these careers were
viewed as masculine, as past studies that applied the concepts of Acker’s
theory of masculinity. According to Acker’s theory, females are less likely
to take STEM classes and have female role models (as cited in Moakely &
Kim, 2014). This may be due the discomfort or lack of confidence females
feel about these fields, resulting in non-STEM career choices and lower
salaries, since STEM fields tend to be higher-paying than traditional caring
professions females tended to enter (Acker, 1990; Allshare & Miller, 2009;
He & Freeman, 2010; Moakley & Kim, 2014).
For the purpose of my study, the levels of STEM classes began in
high school with ninth grade general science, algebra, and computer classes
through 12th grade physics and precalculus, on to postsecondary from
freshman precalculus, to undergraduate calculus, differential equations, to
masters level math. The same levels go for the sciences from basic
freshman biology to higher level undergraduate anatomy and physiology,
chemistry, meteorology, geology, and all other branches of sciences taken
from the undergraduate to the masters level or doctoral level.
Literature Review
History of Gender Gap in STEM fields
Despite recent antidiscrimination legislation, trended towards equal
opportunity and equal pay for females, when it comes to entering the STEM
fields), females still lag behind. According to the Department of Commerce
(2011), less than 25% of STEM positions in America were occupied by
females. Despite their success filling male’s jobs during WWII and
throughout history, today, while females make up 47% of the workplace,
they account for only 20% of STEM careers (Kenney et al., 2012).
According to Milgram (2011), labor statistics from 2005 indicated that only
15% of females were in the field of engineering, 8% in manufacturing,
14.5% in IT, and 9.6% in architecture, and these percentages are of all
workers in each perspective career choice category, masculinising these
professions
The reason that after WWII females did not sustain the STEM
positions they gained during the war is because when the men returned from
the war, the women were told to leave their jobs so that the men would have
employment. It was still believed that a woman’s place was in the home and
that it was the man’s job to provide for the family (Kenney et al., 2012).
Also, females hold a disproportionately low percentage of science and
engineering degrees (Department of Commerce, 2011). Although the place
of females in society has changed in the last 120 years, there are still many
who believe a female’s place is in the home (Kellerman & Rhode, 2007;
Kenney et al., 2012). Many factors have led to this gender gap historically.
Historically, there are many factors that have led to the gender gap in
STEM careers. Some of these factors included the masculinisation of STEM
fields, communicational and behavioral differences between males and
females since childhood (Paciello, Fida, Tramontano, Lupinetti, & Caprara,
2008), the different views on careers between males and females, the lack of
role models that females have in STEM fields (Kenney, McGee, &
Bhatnagar, 2012; Moakely & Kim, 2014), and the lack of confidence
females have in their math and science abilities resulting from societal
stereotypes (Kenney et al, 2012; Moakely & Kim, 2014), creating the need
to study this relationship of STEM classes taken in high school and the
postsecondary level, role models, and career choices along with salaries.
Historically many females may have been reluctant to enter math and
science related fields, take STEM classes, or choose these vocations as a
career because of gender stereotypes, such as in IT fields (He & Freeman,
2010). Some reasons for this reluctance have been the lack of female role
models in the field and not being encouraged to enter STEM fields early in
their academic careers. Kenney et al, (2012), Milgram (2011), and Moakley
& Kim (2014) believed that it was crucial for more females to enter STEM
careers because females bring in a diverse perspective which would broaden
perspectives in a masculinised field.
With their diverse perspective, females have made considerable
strides in legal and medical careers, but have not been as successful
accessing the STEM fields. According to London, Rosenthal and Gonzalez
(2011), despite the recent advancement females have made in non-traditional
careers such as doctors, lawyers and STEM careers, which include scientists,
technical personnel, engineers, and mathematicians, they are still vastly and
pervasively under-represented in STEM fields. This underrepresentation was
an example of a lack of access to STEM careers because of discrimination or
females avoiding science and math classes and careers. Although the
number of females in legal and medical careers has increased, there is a
considerable gender gap between the number of males and females in STEM
careers (London, Rosenthal, & Gonzalez, 2011). In each of these fields,
there is still a wage gap where females earn considerably less because they
choose careers other than high-paying STEM careers (Royal, 1996). This
warrants career management which according to Seibert, Kramer, Holtom
and Pierotti (2013), is about setting intrinsic and extrinsic goals such as a
more challenging career with a higher salary.
There is wage disparity between men and women in STEM fields,
showing that women are still lacking success. London, et al, (2011)
attributed these gaps to negative stereotypes and hoped to research in more
detail as to why these stereotypes persist. London et al, (2011) performed an
Experimental Sampling Method (ESM) which is a method of data collection
also called diary research, recording everyday experience, or conducting
event sampling research where researchers employ repeated measures to
sample behavior, emotions, or experiences, over a period of time or a
particular event. In this case, they measured engagement and success for a
sample of females in STEM fields using surveys and a series of math and
science exams over a period of time. The authors revealed in their findings
that the manipulation of the variables in the study may undermine the female
performance in math and science on these exams in London, et al’s study.
Moreover, these performances may influence whether or not females major
in STEM fields in college or university. Such manipulation may give the
erroneous impression that females do not have the same math and science
abilities as their male counterparts, making it crucial to conduct my
relationship study which is one reason that females have accessed STEM
careers at reduced rates in comparison to males (London et al., 2011).
My Study
In this relationship study I conducted, the research questions on what
kind of relationship there is between STEM classes taken by females and the
role models they have in high school and college and their career choices
and the salaries they make as a result. My study was quantitative, using an
online survey instrument where I allocated responses to a group of career
choice categories where the measure is the number of STEM classes taken
by females. Then ANOVA was conducted to determine which career choice
groups needed the highest average of STEM classes for research questions
one and two. Then I ordered the averages, and conducted a post hoc to
determine which differences are significant. For the predictor variables in
research questions three, I conducted a multiple regression. I have not seen
many studies use these methods, which could be one of the gaps I can fill.
Few authors have focused on how specifically discouragement
(Bouvier & Connors, 2011) from math and science have left females without
the proper technical and leadership training to help to influence career
choices in the STEM careers (Dugan, et al, 2013). This discouragement had
been a factor in the gender gap because of the lack of role models or STEM
courses they take, which has negatively influenced career choices (Muchiri,
Cooksey, Di Milia, & Walumbwa, 2011; Moakley & Kim, 2014). For this
reason, I concentrated on the number of role models as well as the number
of STEM classes taken to see if this lack of role models impacts on career
choices in STEM and if the lack of female STEM role models discourages
females from these career choices. The lack of role female models may be
one of the reasons for the gender gap (Muchiri, Cooksey, Di Milia, &
Walumbwa, 2011). This may be a gap I can fill because I can focus on the
number of role models, where many other studies have not focused, adding
to the
literature.
One possible barrier for females in accessing STEM careers is that
females have the option not to choose STEM careers. Many females may
feel that the courses are too difficult (Klawe, 2013). This concern of
difficulty may be because of the traditional societal belief that females were
not scientifically and technically oriented, as well as the lack of guidance,
support, mentorship, and the lack of course exposure to these careers
(Hensvik, 2014; Kay & Shipman, 2014; Obama, 2014). Traditionally,
females have been discouraged and made to feel that they would be unable
to succeed in math and science classes because they believe the classes are
too difficult, which may correlate with females underestimating their
abilities in STEM tasks (Dunning, Kruger, & Williams, 2013; Hensvik,
2014; Kay & Shipman, 2014; Obama, 2014). Many females who were
proficient in math have been discouraged from STEM and encouraged to go
into accounting and finance (Obama, 2014).
Moreover, according to Obama (2014), many females were
discouraged from STEM classes and those who were considered good in
math, were steered towards finance or accounting instead of STEM because
of the common belief that females are not as good at the sciences and
technology as males and the lack of female role models (Farland-Smith,
2009; (He & Freeman, 2010; Klawe, 2013; Moakley & Kim, 2014).
Furthermore, according to Byers-Winston (2014), women and racial/ethnic
minorities hold less than 25% and 9% of STEM jobs requiring a college
education, respectively, considered underrepresented minorities in STEM
occupation. This underrepresentation of females was a rationale for my
study examining the relationship between the number of female role models
and number of STEM classes taken with career choices and salaries.
Furthermore, Stout, Dasgupta, Hunsinger, and McManus (2011) saw the
option not to take STEM classes as an issue of a low self-concept on the part
of females, because of the belief that STEM classes were too difficult, based
on a series of studies they conducted. It is possible that females may enter
STEM careers at lower rates than their male counterparts because they may
have fewer resources to take the necessary classes, lower self-concept on
their abilities, underestimating their abilities, and may be discouraged from
these classes early in their academic careers. As a result of low self concept,
Stout, et al, (2011) postulated that this may be a reason why females have
been exercising their freedom to avoid STEM careers because of their lack
of confidence in their abilities (Dunning et al, 2013; Hensvik, 2014; Kay &
Shipman, 2014).
Unlike their male counterparts who tended to overestimate their
abilities especially if their abilities were less than stellar, females tended to
underestimate their abilities in science and math even when their abilities
were stellar (Dunning, et al, 2013; Kay & Shipman, 2014). According to the
Dunning-Kruger effect, those who performed well tended to underestimate
their abilities and those who performed poorly tended to overestimate their
abilities (Dunning, et al, 2013). Moreover, males tend to overestimate their
abilities and females tend to underestimate their abilities especially in STEM
abilities (Kay & Shipman, 2014). One remedy to combat this issue would be
to increase the requirements of science, math, and technology in high
schools so that females or any students cannot escape this essential training
(Brown, et al, 2011) in order to help close this gender gap. In contrast,
Kenney, et al. (2012) believed that one way to reduce the gender gap is to
offer females more spatial training. These authors postulated that this kind of
training should occur when females are small children so that they can be on
the same playing field as males and not made to feel any less competent than
their male counterparts. This requirement may boost female confidence in
these fields and encourage females to take more STEM classes which is a
variable to be examined in my study
Like Hensvik (2013), Kellerman and Rhode (2007) saw this gender
gap in the STEM careers as also negatively impacting female access to
leadership and management positions as well as STEM positions. According
to Dugan, Fath, Howes, Lavelle, & Polanin (2013), for female STEM majors
in college had significantly lower leadership efficacy than their male
counterparts. This may occur because parents and teachers may be
discouraging females from majoring in entering these fields or from
becoming leaders in these fields (Dugan, et al, 2013; Farland-Smith, 2009).
According to Farland-Smith (2009), young females lose interest in science
early in life, which may be due to the fact that they are not nurtured in this
area. This lack of interest deters them from pursuing science careers. Also,
in contrast to my hypothesis, Farland-Smith stated that no matter the number
of STEM classes females took, females may still avoid STEM careers if they
perceive them as boring, difficult, or without role models. However,
Farland-Smith also stated that discouraging young females from taking
STEM courses was a contributing factor to resulting in their avoidance of
these fields, which may be revealed from my data collection.
Review of Recent Research on STEM Careers and Females
Similar to Farland-Smith (2009) who said that one reason females
avoid STEM fields is that they believe they are not enjoyable and fun, a
similar study was conducted at the famous Harvey-Mudd College for
computer science, where a similar conclusion was drawn. Klawe (2013)
stated that before she conducted this study and implemented this
programme, there were only 10% of females in computer science in 2006. In
2014, there are 40% and Klawe (2013) conducted a survey on potential and
incoming female students and asked them to give three reasons why they did
not major or have interest in computer science. The three main reasons given
were that this field was boring, not fun or interesting, also revealed by
Farland-Smith (2009). The sampled females did not feel they had the
confidence to do the difficult math (Kay & Shipman, 2014) and thirdly the
field attracted geeks who were isolated from the rest of society and the
community. Upon receiving this input, Klawe created a programme to
encourage an increase in participation by females in computer science which
supports my hypothesis that the more STEM classes and role models
females take and have the more likely they will choose to enter these high
salaried career fields. According to Klawe (2013), as a result of this
programme, there was a 30% increase in the percentage of females who
majored in computer science at this college by making the field more fun
and interesting, offering more math support and finding easier methods of
mastering the math.
Lastly Klawe’s (2013) programme created a teamwork community
similar to what was advocated by Manning (2012) who said that it was not
his legacy, but the team’s legacy that was important. Having a sense of
community and belonging helped these females to feel less isolated and
more encouraged to take computer science and to be part of a larger
scientific community not marginalized by society. Furthermore, this
programme showed that when females are encouraged in math and science,
and science and math are made to be fun, their confidence to succeed in
these areas is increased, which can be applied in public schools (Farland-
Smith, 2009; Milgram, 2011). A sense of community and teamwork as well
as being fun are important to females, therefore, science needs to be made
enjoyable through community interaction and group hands-on
activities.
Raising the math and science requirements for all students in K-12,
may help force female students to access the same math and science training
as their male counterparts. According to Brown, et al. (2011), if schools
require more STEM classes for all students, this additional training will help
all students, which will inadvertently help to reduce the gender gap in STEM
training and the low rate of employment (Glass, Sassler, Levitte, &
Michelmore, 2013).. The problem of increasing interest among females for
STEM careers was of vital interest currently because females have either
been less interested or less confident in their STEM abilities (Dunning, et al
2013; Kay & Shipman, 2014). In addition, one-way to combat this issue is
by increasing the requirements for STEM education, train teachers in this
area. Due to this persistent gender gap, my study is a very timely problem.
Reasons for the Gender Gap in STEM Careers
STEM careers have been traditionally masculinised, attracting mostly males.
Carrell, Page, & West (2010) and Kenney, et al, (2012) postulated that
society masculinised STEM careers much in the same way that Anderson
(2006) discussed that society recently feminized public relations. According
to He and Freeman (2010) and Klawe (2013), society viewed STEM careers
as more geared towards males because of their technical nature. Females
were not generally encouraged to enter technical careers or engage in
technical tasks from childhood (Erikson, 1980; He & Freeman, 2010;
Klawe, 2013).
Societal Factors
In addition to masculinisation, there are some negative stereotypes
about females and their abilities in STEM fields such as that that females,
unlike makes, do not have the innate abilities for STEM and their supposed
reluctance to work long hours (ByersWinston, 2014). These negative
stereotypes of society include beliefs that males are better at math and
science than females, which may make them reluctant to enter these career
fields for fear of failure according to Carrell, Page, and West (2010),
Kenney, et al. (2012), and Stout, et al., (2011). These stereotypes came from
several factors. One factor is the cultural expectations of how females should
behave which in many cultures, both individually and in groups (Jiang,
2010) is passive, and males are expected to be assertive or aggressive,
according to Alshare and Miller (2009). STEM careers tended to attract
persons who were assertive, goal-oriented, focused on career goals, and who
are aggressive in achieving these goals (Allshare & Miller, 2009; Carell, et
al., 2010; He & Freeman, 2010). Alshare and Miller also postulated that
because of this expectation in behavior, many females avoided STEM
courses in school and careers because these careers were viewed as
masculine. Also, according to He and Freeman (2010), many females were
thought to be technically inferior in their abilities to males because they
were never encouraged to develop these skills because they were never
encouraged to develop these skills. Moreover, many females had low
confidence in these abilities (Klawe, 2013). This may have contributed to
feminine avoidance of these careers, making my relationship study very
crucial.
There is even a lack of female STEM professors. According to
Carrell, Page, & West, (2010), one major factor to the gender gap between
male and females in STEM careers was due to the lack of female professors
in their courses. This relates to the lack of role models that females can
emulate in these careers, which is one reason why females felt the freedom
not to choose STEM careers (Stout, Dasgupta, Hunsinger, & McManus,
2011). Female professors earn considerably less than male professors and
they only hold 25% of post-doc fellowships in STEM fields and only 39% of
STEM faculty posts (London, et al, 2011). The broader question may be the
role of the professor’s gender and prediction of STEM careers (Kenney, et
al., 2012; Stout et al, 2011). The same issue was found in high schools as
well, where again most of the science and math teachers were males
(Carrell, et al. 2010).
In addition to the role of the professor and lack of role models,
another reason for the gender gap in STEM careers was that a large
percentage of females have had a lower opinion of their ability to perform in
math and science than males. Correll (2004) conducted an experiment of
males and females who were told they were being tested for a college
admission test. The experiment was an evaluation of a model where
respondents were asked to conduct a math related task and made to believe
females are better at this task. Correll (2004) also examined the role of
culture and what it played in how females feel about their math and science
abilities (Byers-Winston, 2014; Sikora & Pokropek, (2012). Correll claimed
that in the prevailing culture, math is seen as masculine by both sexes and
therefore, females tended to avoid math classes and career choices, making
my relationship study very important. As a result of the experiment, Correll
concluded that because females rated their aptitudes in math, lower than that
of their male counterparts, they were less likely to enter STEM careers
because they require a great deal of math. This was because females were
made to feel they were less competent than males in these technical skills in
cultures across the board, according to Correll and He and Freeman (2010).
However, when females were told they performed better than makes, their
aptitude improved, showing that encouragement may help females perform
better. According to Carrell et al. (2010), preparedness and aptitude seem
similar for both genders and does not predict access to STEM careers and
employment. Carrrell et al. (2010) believed that the gender gap is due to lack
of preparedness by females meaning lack of training in STEM, resulting in
the low rate of employment for females (Glass, Sassler, Levitte, &
Michelmore, 2013)., which may be due to lack of encouragement but not
lack of aptitude.
Despite societal stereotypes that females are not as technically
minded, when several authors conducted studies comparing males and
females in math and science tests, many times the females outperformed the
males. For example, similar to Correll (2004), who compared the math
performances between the sexes on exams, Kenney,
McGee, and Bhatnagar (2012) compared the differences between the sexes
in math. They asserted that historically females outperformed males in the
math classes, but the males outperformed the females in high stake
university entrance exams like the SATs, which contributed to why females
avoid STEM careers (Kenney, et al, , 2012), creating a need for my
relationship study. Kenney, et al (2012) also stated that females were
outperformed in spatial skills. This was because females were made to feel
they were less competent than males in these mathematical, scientific,
technical and spatial skills in cultures across the board, according to Correll.
Sikora and Pokropek (2012) also agreed that females were made to feel less
competent than males across diverse cultures, similar to Correll. Therefore,
females have lower self-esteem and a lower opinion of their math abilities
which hinders them and creates a gender gap, further discouraging females
from entering STEM careers, creating a wider gender gap (Kenney, et al,
2012). Furthermore, females have not been encouraged to partake in STEM
activities, classes, and careers and have had less spatial or mechanical
training than their male counterparts (Correll, 2004; Sikora & Pokropek,
2012), due to discrimination and societal attitudes.
Discrimination and the pervasive attitudes in society is that females
have less spatial, technical, and mechanical abilities than males. This is a
major reason that females do not see themselves as capable of math and
science is due to discrimination and societal attitudes (Milgram, 2011).
These prevailing attitudes say math and science are masculine careers and
males are better at spatial and mathematical tasks than females. These
generalizations were simply not true, as females performed just as well as
males as cited by Carrell, et al, (2010); Kenney, et al, (2012). According to
Farland-Smith (2009), that from the time many females are young children,
they have a lower perception of their math and science abilities. In
agreement with Farland-Smith, Watt, et al. (2012) sampled three groups of
high school students in the United States, Canada, and Australia and
employed a multivariate analysis of variance (MANOVA), which revealed
differences in early gender socialization as having an impact on courses
taken in high school and preliminary career choices (Watt, et al., 2012). She
claimed that because females were not encouraged in math and science play
as children, and in their socialization, this may impact their interest in taking
math and science classes in high school and college. This may be a
contributing factor as to why some females avoid STEM classes, majors,
and careers. This avoidance may stem from early socialization, according to
Watt, et al, 2012).
Females have sometimes been discouraged from spatial, technical,
mechanical, math, and science activities in their early socialization. Watt, et
al. (2012) also postulated that this early socialization may relate to how
females rate their own abilities in math and science. Hence, females may
avoid restrictive math and science courses required to enter a university. For
this reason, according to Watt, et al., many females avoid these STEM
majors when entering college and choose a social science instead because of
their lower perception of their own abilities to be successful in these courses,
creating a need for my relationship study.
Like Correll (2004) and Sikora and Pokropek (2012), He and Freeman
(2010) also agreed that females thought of themselves as less technically
minded because of these same attitudes perpetrated by society. Yet if
thought of us competent, the females performed comparable to their male
counterparts. These social attitudes were also the same in management,
according to Tallon-Hammill, (2010). In addition, according to He and
Freeman (2010), society also believed that females were less technically
minded and less competent at technical fields such as IT, which is among
some of the stereotypes about females. The salient cultural stereotype
perpetrated about females is that they are not as good at math and science,
or as technical minded as males or as acclimated as males towards fields
such as IT (Buche, & Scillitoe, 2007; He & Freeman, 2010). According to
Buche and Scillitoe (2007), these attitudes and traditional beliefs that
females were not as technically minded as their male counterparts, begin in
childhood. These authors postulated that in childhood, females are
encouraged in caring forms of child play in motherly, caring, comparison,
and cooperative roles as claimed also by Aldwin (2009), Noddings (1986)
and Sherblom (2008). Whereas, males as children are encouraged in
competitive, scientific, and technical play. Buche and Scillitoe believed that
this early play may also be a factor in why females may not believe they are
as competent in the sciences as their male counterparts. Furthermore,
females may not be experiencing as much early exposure to math and
science in their play and early socialization as males (Brown & Tappan,
2008; Erikson, 1980, 1997). These differences since childhood contributed
to the gender gap in STEM careers which needs to be addressed. I hoped to
better educate colleagues, students, faculty, business, researchers, society,
and the general public on how to best understand the societal factors
leading to this gender gap which creates a greater need for my relationship
study on the relationship between STEM courses taken by females and
career choices. All of these parties have the power to bring about social
change through narrowing the gender gap in accessing STEM careers.
Educational Factors
In addition to societal factors, educational factors play an important
role in the gender gap. For example, it was the lack of STEM classes taken
by females and the lack of same sex STEM role models that influenced
female avoidance in STEM careers. Females were discouraged from such
employment and careers. Furthermore, math and science college prep course
differences such as SAT prep are also not strong predictors of gender
differences in university majors, according to Carell, et al (2010). The SAT
prep courses and exams are the same for both genders. In contrast, Kenney,
et al. (2012) believed that one-way to reduce the gender gap is to offer
females more spatial training. These authors postulated that this kind of
training should occur when females are small children so that they can be on
the same playing field as males and not made to feel any less competent than
their male counterparts. London, et al (2011) and Sikora and Pokropek
(2012) also agreed with Correll that females were made to feel less
competent than males across cultures, particularly in math and science. This
inferiority complex has taken place in school and in society in general,
which concurred with the developmental theoretical construct and
conceptual framework of Erikson (1971). This complex has given many
females the excuse or choice not to even try to enter these fields (Kay &
Shipman, 2014).
How students rate their math and science abilities, impacts on their
performance in these math and science areas. Furthermore, when students
believed their performance was low in these STEM areas, they avoided
taking these classes. Like Watt, et.al (2012), Correll (2004) affirmed that the
higher students rated their own math ability, the likelier they were to take
classes in math and choose a college major or career in math. In addition the
findings also concluded that math abilities were associated with masculinity
(Kenney, et al, 2012). Likewise, Farland-Smith (2009) asserted that females
saw themselves as less competent in math and science and had a lower
perception of their abilities than their male counterparts. This lower
perception resulted in females avoiding taking math and science classes that
were not required and caused them to avoid these careers (Stephens, 2004;
Watt, et al, 2012). However, Kenney, et al, (2012) stated that although
females rated their abilities lower than males, they outperformed males in
math classes in high school but the males outperformed the females on the
SATs and entrance exams. Kenney, et al, (2012) and Carrell, et al (2010)
believed that females have no difference in math abilities than males. Thus,
similar to Correll (2004) and Farland-Smith
(2009), Kenny et al. postulated that it was society’s attitudes towards female
abilities in science that brought about the idea that females are not as
proficient in math and science as males. These attitudes have helped to bring
about a gender gap in STEM careers, creating a need for my relationship
study.
These attitudes resulted in research comparing the difference between
the sexes in math and science proficiency. Similar to Correll (2004),
Kenney, et al (2012) compared the differences between the sexes. They
asserted that historically females outperformed males in the math classes,
but the males outperformed the females in high stake university entrance
exams like the SATs, which contributed to why females avoid STEM
careers (Kenney, McGee, and Bhatnagar, 2012). Kenney, et al (2012) also
stated that females were outperformed in spatial skills. This was because
females were made to feel they were less competent than males in these
mathematical, scientific, technical and spatial skills in cultures across the
board, according to Correll. Sikora and Pokropek (2012) also agreed that
females were made to feel less competent than males across cultures. As a
result, some females have lower self-esteem and a lower opinion of their
math abilities which hinders them and creates a gender gap, further
discouraging females from entering STEM careers, creating a wider gender
gap (Kenney, et al, 2012).
This theoretical construct of Erikson’s developmental stages stated
that what happens early in life impacts what happens later in life, generally
in the young adult or intimacy stage and the midlife or the generativity stage,
in the sixth or seventh life stages of Erikson’s life stages of development
(Erikson, 1980). If females are not encouraged in math and science at a
young age, they tended not to have interest later in life (Erikson, 1971; 1980;
He & Freeman, 2010). Sikora and Pokropek (2012) also agreed that females
were made to feel less competent than males across cultures. Consequently,
females have lower self-esteem and a lower opinion of their math abilities
which hinders them and creates a gender gap, further discouraging females
from entering STEM careers, creating a wider gender gap, negatively
impacting their choices of courses and careers. One of the objectives of my
study was to educate the scholarly world and the public on these
discouraging barriers, hoping to encourage more females to take STEM
classes, creating a need for my relationship study.
Professional/Career Factors
There were many factors that have resulted in the gender gap in
STEM careers, discouraging females from entering science and technical
careers. These careers have been traditionally labeled as male. Acker (1990)
and Royal (2007) both postulated that when females worked in jobs that are
traditionally male, the self-image and self-esteem of these females became at
odds with what they do naturally to perform, which are more caring
professions, according to Noddings (1986). These females were expected to
act the same way that a male would in the same role (Acker, 1990). In other
words, since their school days, a large percentage of girls across cultures
were taught to be submissive, quiet, and “good”. They would perform well
in their grades, but they did not learn assertiveness or competitiveness,
necessary to access higher paying careers in STEM, leadership, or
management (Dunning, et al 2013; Hensvik, 2014; Kay & Shipman, 2014).
These expectations have also contributed to discouraging females from
majoring in STEM or technical degrees and choosing these careers.
Females who did attain STEM degrees, particularly in computer
science, and engineering, did not ascend into the upper management in those
fields, hence the glass ceiling (Kellerman & Rhode, 2007; Noddings, 1986).
Furthermore, the freedom and discouragement not to take STEM classes has
resulted in females taking fewer of these classes than their male
counterparts. Some females may therefore, question their abilities in these
technically minded fields, reluctant to try to enter these fields (Kay &
Shipman, 2014). Furthermore, females have not been encouraged to partake
in STEM activities, classes, and careers and have had less spatial training
than their male counterparts (Correll, 2004; Sikora & Pokropek, 2012).
Since society views spatial, mechanical, technical, and science
abilities as masculine, females also viewed these abilities as masculine,
resulting in less training and experience in these fields for females.
According to He and Freeman (2010), society considered technical, spatial,
and IT career fields more suitable to males than females. Moreover, Correll
(2004) also asserted that there is a prevailing culture about math that it is a
masculine subject and females tended to avoid math in class or as a career.
Similarly, Farland-Smith (2009) asserted that females saw themselves as less
competent in math and science and had a lower perception of their abilities
than their male counterparts. This lower perception resulted in females
avoiding taking math and science classes that were not required and caused
them to avoid these careers (Watt, et al, 2012). However, Kenney, et al.
(2012) stated that although females rated their abilities lower than males,
they outperformed males in math classes in high school but the males
outperformed the females on the SATs and entrance exams. Kenney, et al
(2012) believed that females have no difference in math abilities than males,
but that it was their lower perception of their abilities, and the masculine
bias of these standardized tests, which deterred them from entering STEM
careers, creating a need for my relationship study. Thus, similar to
Correll (2004) and Farland-Smith (2009), they postulated that it was
society’s attitudes towards female abilities in science that brought about the
idea that females are not as proficient in math and science as males. These
attitudes have originated from traditional stereotypes that females are better
at caring, soft, humanitarian professions, whereas STEM careers have been
masculinised by society, according to Kenney, et al (2012) and Sikora &
Pokropek (2012). These attitudes resulted in females having a lower
selfconcept in these STEM abilities.
As a result of this lower self concept in their STEM abilities, females
may not have developed the same technical background in STEM classes
that their male counterparts had. It is possible that these females have not
taken the same amount of technical courses (He & Freeman, 2010). Thus,
similar to Correll (2004) and Farland-
Smith (2009), Kenny, et al (2012) postulated that it was society’s attitudes
towards female abilities in science that brought about the idea that females
are not as proficient in math and science as males. Females appeared to
avoid taking STEM classes in high school that were not required and
females did not take as many STEM classes and were not as encouraged to
take these classes by their parents and teachers as their male counterparts
(Farland-Smith, 2009; Milgram, 2011).
Even today, females are still underrepresented in the STEM fields.
According to London, et al (2011), despite the recent advancement females
have made in nontraditional careers such as doctors, lawyers and especially
STEM careers they are still vastly and pervasively under-represented.
Therefore, only 18% of females major in engineering in a college or
university. Watt, et al (2012) postulated that many females avoid taking
math and science in high school because they believed these courses are too
difficult, which may contribute to the reason that females seemed to perform
below males on the SAT math section in that particular study, according to
Kenney, et al (2012). The reason for scoring lower than males may have
been because females lacked confidence in math and science because of
being told they were not as acclimated to these courses as their male
counterparts, and males were more encouraged in these skills and courses
(Correll, 2004). In order to improve the self-concept of females in the areas
of math and science, Stout, et al, (2011), believed that the freedom not to
choose such careers was fueled by societal stereotypes about the abilities of
females in these fields as well as the low number of female role models or
experts to emulate (Kenney, et al (2012), necessitating my relationship
study.
Another major issue that females face in the barriers of entering
STEM careers was discrimination and the masculinisation of these careers
(Acker, 1990). As a result of these factors, these careers have been gender
segregated (Farland-Smith, 2009; Sikora & Pokropek, 2012). Traditionally,
females have entered non-science, caring professions and males have been
entering science professions as doctors, medical professionals, scientists,
researchers, astronauts, geologists, biologists, and other professions in the
hard sciences. In fact, even within the sciences, Milgram (2011) examined
the horizontal gender gap in tertiary education when it comes to STEM
careers and who because of segregation males preferred computers,
engineering, or math (CEM) and the physical sciences and females preferred
biology, or living systems, agriculture, photonics, or health (BAH). One
reason females may prefer photonics and optics are that these applications
are used in caring professions like healthcare (Milgram, 2011; Noddings,
1986).
Male dominance in the computers, engineering, and math, resulted in
a limited perspective. This lack of diversity, with males dominating the
CEM fields as concluded by Milgram (2011) can be very limiting in scope,
viewpoint, perspective, and outlook.
This was consistent across cultures. There has been a slight increase in
females entering STEM careers but mostly on, the BAH careers (Milgram;
2011). However, according to labor statistics from 2005, 15% of females
were in the field of engineering, 8% in manufacturing, 14.5% in IT, and
9.6% in architecture (Milgram, 2011). The problem may be that in high
school, females do not take as many STEM classes as their male
counterparts. These females may only take what is required to graduate.
Furthermore, females have not been as encouraged as their male
counterparts to take STEM classes and choose these careers (Milgram,
2011). Therefore, only 18% of females majored in engineering in a college
or university as of 2005, according to Sikora and Pokropek, (2012) and
Milgram (2011). No female in my sample population majored in or chose a
career in the field of engineering which went in accord to the findings of
Sikora and
Pokropek and Milgram.
It can be very intimidating for females when they are the only female
in a class of all males. Therefore, it is necessary to conduct some outreach to
actively encourage and recruit females for enrollment in STEM classes,
majors, and careers from high school to college to graduate school and the
work place. For this reason, I asked about the number of STEM classes
taken since high school by the female participants in my study from the
alumni associations of four Long Island universities, and the few Walden
participants added due to the small sample, and their career choices as well
as salaries as they compared with their male counterparts.
As I hypothesized, the more STEM classes females take, the better
they would perform in these fields. Correll (2004) conducted an experiment
designed to evaluate the hypothesis that if students thought their abilities in
STEM courses, tests, and skills were proficient, they would perform better
and the gender gap in these courses and their majors and career choice would
decreases. Therefore, she used a probability sample of high school and
college students and measured the degree to which cultural thoughts about
gender and math played a role in career choices. The experiment was an
evaluation of a model where respondents are asked to conduct a math related
task and made to believe males are better at this task. The students studied
were brought into the lab individually and told they were pre-testing for a
national admissions exam, completing several computer tests using a
contrast sensitivity scale of 100 items evaluating tasks on their masculinity,
using a one-way ANOVA. I used an ANOVA to quantify career choices.
One group in this study by Correll was made to feel that males were better at
this task and the other group was made to feel that both genders were equally
proficient at this task. Career choices and gender segregation in academic
activities seem to begin as early as high school and continue into college;
according to Correll (2004). The gender belief associated with a task was an
independent variable and self assessment is the dependent variable. Correll
then stated that if males and females make different assessments of their
abilities, this will impact the career paths they will take. The findings in this
study demonstrated that males rated their abilities at this task higher than
females and that the group that was made to feel the males were better at the
task, rated males higher due to these cultural thoughts, also postulated by He
and Freeman (2010). The higher students rated their own math ability, the
likelier they were to take classes in math and choose a college major or
career in math.
If societal attitudes would view STEM careers as gender neutral, the
gender gap would narrow. This was evident from the findings from this
study by Correll (2004) which she also concluded that math abilities were
associated with masculinity, which
Farland-Smith, (2009), Sheaffer, Bogler, and Sarfaty (2011) and Sikora and
Pokropek,
(2012) also agreed with this finding about the masculinity of STEM fields
and careers. According to Correll (2004), the group that was made to feel
that both genders were equally proficient saw little or no gender difference
in assessment and evaluation of the task. These results were compared to a
former study that showed that males rated their math abilities higher than
their female counterparts rated their math abilities.
If the general public had more faith in the abilities of females in
STEM fields, and reduce the masculinity associated with these fields,
females would show improved performance and interest in these fields.
Kenney, et al (2012) and Stout et al (2011) believed the reasons for females
avoiding STEM careers were because females are discriminated against and
historically science careers are associated with males and masculinity and
not females and femininity, and because of perceptions of the differences in
ability based on gender and career choices. Females faced discrimination in
reference to their abilities in these STEM fields and have been discouraged
from entering these fields early in their academic careers (Erikson, 1971;
1980; Stout, et al (2011). Stout, et al (2011) postulated that males were
deemed superior in math and science, discouraging females from entering
these fields. According to the Department of Commerce (2011), females
only hold 24% of STEM jobs and careers. One reason could be the lack of
female role models.
If females had more role models to inspire them in youth, to enter
STEM careers, and take these classes, perhaps, they would be inspired to do
so. To demonstrate this trend, Farland-Smith (2009) conducted a mixed
methods study (Tashakkori, & Teddlie, 1998) on the attitudes of middle
school 26 females at a Midwestern university as a result of their experience
at a science camp called Side by Side. Here the females experienced what is
was like to work side by side with various scientists as role models in fields
such as biology, anthropology, physics, chemistry, and biology. The authors
wanted the female participants to see science as fun. However, this study did
not ask about their courses or whether or not they wanted to enter the STEM
field, but just their attitudes as a result of the experience, particularly their
experience with the role models. This was a gap I filled with my study. Also,
this was a middle school population and my study was adults who graduated
high school and college or university. The idea was that if females had an
enjoyable experience, with inspiring role models, seeing science as fun, their
interest would increase.
Same sex role models may inspire females to increase their perception
about their math and science abilities. To increase females’ self-perception
about their math and science abilities, Stout, et al (2011) concluded that role
models and same sex experts may inoculate stereotypes and increase the self
concept of females to entre STEM careers, reducing the gender gap
(Kenney, et al, 2012). Therefore, Kenney at al., London, et al and Stout, et al
agreed that if females were encouraged with success, and had same sex
experts as role models, females would be encouraged to take math and
science in both secondary and tertiary education (London, Rosenthal, &
Gonzalez, 2011; Stout, et al ;2012). Furthermore, because of the lack of
same sex role models for females in STEM careers, females have avoided
these careers thinking they are too difficult as postulated similarly by
London, Rosenthal, & Gonzalez, (2011) and Stout, et al, (2012).
Developmental Factors-Impact of Early Development
Academic and social exposure early in development impacts choices
made later on in life. Erikson (1980) emphasized that the academic choices
made in youth impacts on the career choices and salaries later on. Noddings
(1986) also emphasized that females tended to choose caring professions
instead of scientific ones, which tended to pay less than STEM fields. Even
as children, females are encouraged in these caring roles in their play with
dolls. Females are not encouraged to play or tinker in the sciences,
technology, or in building things like their male counterparts (Erikson, 1980;
Noddings, 1986). According to Watt, et al (2011), there was only a 9%
participation rate in STEM careers in Anglo nations such as Australia
Canada, and the US because both males and females see STEM classes as
difficult. However, the participation for females is lower than males because
females avoid these careers believed to be too intense, and therefore enter a
social science field like law or political science or healthcare. Females need
encouragement by role models to enter these fields.
Differences Between Males and Females in Career Choices
Females choose careers for different reasons than males which makes
my relationship study a necessity. Correll (2004) asserted that males choose
careers they enjoy and are good at and females tend to choose careers that
are flexible and a balance between work and family. Farland-Smith (2009)
and Milgram (2011) postulated that many young women are reluctant to
sacrifice their personal and family lives in pursuit of their careers. This may
contribute to why females do not have the same informal career networks
that males have which help to advance one’s career. This applies to
leadership, or STEM careers (Farland-Smith, 2009; Milgram, 2011).
Moreover, in STEM careers, and positions of leadership, married males are
perceived as responsible, and married females appear to be perceived as
someone who will not be dedicated to their careers. This has also put
females at a disadvantage in many career fields including STEM careers
(Correll, 2004; Farland-Smith, 2009; Milgram, 2011).
Lack of role models and lack of encouragement may correlate with
lower interest in science and math at a young age. In addition, Farland-Smith
(2009) stated that females lose interest in science at a young age due to lack
of same sex role models, lack of encouragement, and they believe these
fields are too difficult, boring, and inflexible when it comes to work-family
balancing. Also, Farland-Smith (2009) stated that unless females see science
as fun, they will lose interest very quickly, which was why they created the
reason for their science camp. Furthermore, they wanted to know they will
be able to achieve a good balance of family life and careers, which may
cause them to avoid STEM careers (Farland-Smith, 2009; Kay & Shipman,
2014).
There may be an issue of inflexibility when it comes to balancing
work and family which also may deter females from STEM careers. STEM
careers, according to Correll, are not as flexible a balance between work and
family (Kenney, et al, 2012). Generally part time, caring professions that
offer lower salaries than STEM careers tend to fit this criterion of a flexible
balance between work and family (Gilligan, 1986, 1988, 2008; Noddings,
1986). In addition, since males felt confident in their math and science
abilities, according to Correll (2004), males were more likely to choose to
enter these careers more than their female counterparts. Moreover, females
tended to put their families over their careers and because STEM careers are
not as flexible, these careers are the most difficult to balance with family
responsibilities (Correll, 2004). Also, females communicate differently than
males (Kellerman & Rhode, 2007).
Females generally communicate more passively than males.
According to Brown and Tappan (2008) and Kellerman and Rhode (2007),
females communicate with a more passive voice. According to Gilligan
(1986), females have a different voice, one of an ethic of caring, with a
desire to be nurturing. This may be why, according to Sikora and Pokropek
(2012), females prefer the physical sciences such as agriculture, biology,
living systems, medical, or health to the technical sciences such as
computers, math, and engineering, the latter are perceived as more
masculine. Therefore, because females communicate differently (Argyris,
1991, 2003), they also behave differently, making them more passive, and
nurturing, which is why home health aides, healthcare, or the physical
sciences are the careers of choice for females as opposed to the careers in the
hard sciences.
In addition, females have historically been discouraged from entering
STEM careers due to the external environment which creates societal
discrimination and masculinisation of STEM careers (Acker, 1990). For
example, these careers are not as flexible and family friendly as the caring
professions (Kenney, et al, 2012). Recently, there has been an increase in
females entering these professions, but there was still a large gap because of
the continued discrimination and segregation in the employment world
(Sikora, & Pokropek, 2012). Some of this segregation may originate from
the gender gaps in high school and the younger grades.
In high school, many females take fewer STEM classes and have
sometimes performed less proficient than their male counterparts on the
math portion of the Scholastic Achievement Tests (SATs). According to
Carrell, et al (2010), there is a small gender gap in achievement tests in high
school math and science which is not due to differences in abilities, but
rather differences in self perception and course training (Cheng, Shui-fong,
& Chan, J. 2008, Sikora & Pokropek, 2012). Furthermore, preparedness and
aptitude seem similar for both genders and does not predict access to STEM
careers (Carrell, et al, 2010). Yet Carrell, et al. postulated that this factor is
not a strong predictor of the higher likelihood of males to enter STEM
careers over females. It was the general attitude about these careers as
masculine and the idea that females were not as competent in these technical
careers, as well as the lack of expert role models for females to emulate
(Carrell, et al, 2010; Stout, et al., 2011).
Another reason that females avoided STEM careers is due to some
negative attitudes of female students toward school science. Some of these
negative attitudes may begin in middle or high school (Farland-Smith,
2009). These attitudes originated and were reinforced from several sources,
including the failure of parents to encourage their daughters to enroll in
advanced science courses or pursue scientific careers in middle, high school
and tertiary education. In addition, societal norms govern the
appropriateness of career selection by discouraging females from the
sciences and emplacing the masculinity of science (Farland-Smith, 2009).
This discouragement may also have originated from the time females are
small children when unlike boys who are encouraged to play with science
related items like gyroscopes, and chemistry sets, females were traditionally
encouraged to engage in motherly and domestic roles with dolls, as opposed
to science activities as postulated by Irby and Brown (2011), Kenney, et al
(2012) and Noddings (1986), which stemmed from Erikson’s (1980) social
development theory. Furthermore, young females as children, may view
science as if it is for boys, too difficult, or too boring (Farland-Smith, 2009;
Kay & Shipman, 2014; Klawe, 2013). Farland-Smith (2009) recommended
to young females that in order to break down these barriers, they must find
science role models of their gender. The female students, who participated in
the science camp, had a more positive view of science than those who did
not participate in the Side by Side science camp, where they are exposed to
female science role models (Farland-Smith, 2009).
In addition to seeing the positive influence of female role models and
an increase in the participation in science programmes that are fun, perhaps
as the amount of female role models increase, the idea of women in science
will become a societal norm. Societal norms govern the appropriateness of
career selection to gender. This segregation of career selection was
accomplished by discouraging females from the sciences and emplacing the
masculinity of science (Kenney, et al, 2012), making my relationship study
necessary. The masculinisation of STEM careers was very similar to
Anderson (2006) who discussed the feminization of public relations. There
was a time where public relations was male dominated but now it is female
dominated, however, it is still male dominated at the management level. The
same went for the emphasis on the masculinity of the sciences. This gender
bias discourages females from entering fields. For this reason, Farland-
Smith (2009) conducted a study where a science camp known as Side by
Side with Scientists was designed to encourage females to enter the sciences.
Side by Side with Scientists was a science camp where one can study
how females obtain their perception of science, engaging them and making
science fun. This camp accomplished this by using role models and fun
activities to encourage females to enter STEM careers (Farland-Smith,
2009). Farland-Smith also postulated that many young girls are more apt to
like being scientists if they viewed them as fun and humorous. Teachers and
professors, who are female, played an important role in whether they are
boring or fun which would ultimately grab young females' attention. In
addition, Milgram (2011) stated that having role models that are similar to
young female students play a key role in young women decision's whether to
go into the science, math, or technology field.
When females were inspired by role models performing activities that
are fun, they begin to gain interest in science. At this camp, Side by Side,
these similarities in same-sex role models discussed by Farland-Smith
(2009) and Milgram (2011) were evident in how females reacted to same sex
role models. Female participants worked side by side with scientists to learn
what they do on their jobs daily and the young females gravitated towards
those scientists who were fun and like them. This science camp was
established to create a transformative experience for 26 young female
students to broaden their perceptions and understanding about scientists and
their job functions, conducted at a mid-western university. These
perceptions included where scientists work and the type of work they do. At
this camp, these female students explored biology, anthropology, physics,
chemistry, and biology, where these students were encouraged to conduct a
scientific investigation in these areas of science. This kind of camp helped to
encourage the students when they had role models and saw that they
themselves can succeed in the sciences (Farland-Smith, 2009). However,
despite this successful programme, there is still a gender gap in STEM
careers.
Possible Remedies or Proposed Changes to Reduce Gender Gap
One-way to reduce this gender gap was to connect females with
programmes like Side by Side or networking organizations in the sciences
where females can be exposed to female scientists as role models. Similar to
the programme Side by Side by FarlandSmith (2009), there is also an
organization called WISTEE Connect, founded by Dr. Qian (2013). This
organization connects females in science, technology, engineering, and
entrepreneurship. Qian is also a tenure track associated professor at the
Center for Imaging Science at the Rochester Institute of Technology.
According to Qian (2013), there were still too few females in STEM careers,
advancing to high levels in private industry or academia. According to
Farland-Smith, many reasons why females have not entered these careers
have been the lack of role models and mentors. One-way to rectify this was
to create organizations that can connect females to role models and mentors.
This was the objective of Qian’s organization WISTEE. This organization
provided mentorship, connectivity, and leadership opportunities to females
who aspired to succeed in STEM careers and entrepreneurship.
Teamwork is one way to encourage females to enter STEM careers
and reduce the gender gap because mentorship and teamwork attract
feminine interest (Klawe, 2013). According to Kellerman and Rhode (2007)
many females lack leadership/management opportunities, and this
organization provides opportunities in leadership, entrepreneurship, and
teamwork (Jiang, 2010; Manning, 2012), the latter, which offers training that
helps females work interdependently (Qian, 2013). This organization was
also a place for females to connect with additional mentors and role models
to help them access and advance and be guided into STEM careers and
reduce the gender gap. This study used role models, which is one of my
independent variables in my Research Question 3 and a dependent variable
in my first two research questions because I stated in the hypothesis that the
more female STEM role models a female has, this would positively
influence her to make STEM career choices, mentioned in my first two
research questions, and these careers tend to offer high salaries, a dependent
variable used in Research Question 3 in my study. In retrospect, if a female
chose a STEM career, she would be more likely to have taken more STEM
classes and have had more same-sex STEM role models than females who
chose a non-STEM career.
There were some other reasons for the gender gap in STEM fields.
Like FarlandSmith (2009), London, et al (2011) examined reasons for the
gender gap in the sciences and STEM careers. According to London, et al
(2011), and despite the recent advancement females have made in non-
traditional careers such as doctors, lawyers and STEM careers, they were
still pervasively under-represented. In each of these fields, there is still a
wage gap where females earned considerably less. Females only hold 25%
of post-doc fellowships in STEM fields and only 39% of STEM faculty
posts (London, et al (2011), creating a gender gap in these academic fields
and salaries.
In addition, there are a low percentage of females entering technical,
scientific, and engineering careers. There is a wide wage gap, in business as
well as academia within STEM careers. According to Milgram (2011), who
cited labor statistics from 2005, 15% of females were in the field of
engineering, 8% in manufacturing, 14.5% in IT, and 9.6% in architecture.
These figures are the most updated, which are still somewhat old, which was
a challenge for me when I conducted my research. Some of the reasons for
this persistent gender gap included the reasons mentioned above in this
section which included lack of female role models and professors in these
fields, lack of encouragement by parents, teachers, and society,
discrimination and the pervasive attitudes that STEM careers are masculine
and males are better at math and science, and the lower self concept that
females have in their abilities as a result of these attitudes (Carrell, et al,
2010; Farland-Smith, 2009; London, et al, 2011). These attitudes must be
dispelled and society and the educational system need to realize females are
just as competent as males in math and science, and they too needed to be
encouraged to enter and advance in these lucrative fields.
Review of Methodology
There were several researchers who studied the gender gap in STEM
careers between males and females. Many of these researchers have used
surveys and an ANOVA or regression to analyze the data, which were the
same methods I employed (Achen, 1982; Iverson & Norpoth, 1987). A
regression is very flexible and can be used with many quantitative research
methods such as experiments (Achen, 1982; Campbell & Stanley, 1963),
surveys, including marketing surveys as well as traditional social science
research surveys, and observational research. For the data collection of my
study, I employed an online survey with a5-point Likert Scale and analyzing
the data with a regression where I assessed how close the relationship among
the variables is. I conducted an ANOVA to quantify career choices for this
relationship study between the STEM courses females take, the number of
role models and career choices and the masculinisation of STEM careers
(Iverson & Norpoth, 1987). I used these same variables with salaries as
opposed to career choices.
There were some researchers that used a methodology of data
collection which I employed in my study which is a survey with a5-point
Likert Scale. For example, Alshare and Miller (2009) studied sex traits of
both males and females and employed a survey with a 5-point Likert Scale
to collect the data. According to Alshare and Miller, the traits of masculinity
included individualist, material success, focused on material success. Males
were seen as authoritative, individualist, and assertive. Females were seen as
submissive, and collective, meaning concerned for society. This may
rationalize one reason why females may not choose STEM careers as
postulated by Carrell, et al (2010), Kenney, et al (2010) and London, et al
(2011). This methodology used was valid and reliable for this study, which
was quantitative. Therefore, this survey method worked well for my study
because it was a relationship study examining a relationship between the
number of STEM courses taken in high school and postsecondary education
by my sample of females from the four chosen universities in Long Island
and their career choices. According to McCullough (2011), when
distributing surveys, it was best not to use money to increase response rates
for an academic survey as it may induce cheating and bias, reducing the
validity of the study.
Like Alshare and Miller (2009), Sheaffer, Bogler, and Sarfaty (2011)
also employed a 5-point Likert Scale which is valid and reliable. With this
scale, they tested whether or not masculinity affected and predicted how
prepared one was for preparing for an emergency situation. They found that
masculine traits like assertiveness and authority helped preparation in
emergency situations more than passive traits exhibited by females. Using
the survey, this may have also strengthened the argument on why some
females avoid STEM careers because of their masculinity. Again, this
helped to rationalize the use of a survey method for this kind of study. For
this study which examines the relationship between STEM courses taken by
females and their career choices and salaries, this method is the best form of
data collection (Shao, 2002).Chang and Chuang (2012) employed a
relationship study similar to mine and used an online survey method taking
advantage of a low cost, global method, which was also anonymous (Ahern,
2005).
Here was a rational for choosing the method I have chosen which was
the survey method. The survey design which was quasi experimental,
according to Campbell and Stanley (1963), provided a quantitative
description of trends, attitudes or opinions of a population by studying a
sample of the population. From sample results, the researcher may
generalize about the population (Kalton, 1983. Furthermore, here were some
advantages of online surveys including less keypunch errors, cheaper, more
global reach, greatly reduced interviewer bias, greater interviewer control
over randomization, allowed for customization by the researcher, executive
skip patterns, and logic checks (McCullough, 2011). Furthermore, Chang
and Chuang (2012) used a random stratified sample of first, second, and
third year students, similar to my use of the alumni from four universities in
Long Island, mostly females, born after 1980 with close to 6 years of work
experience. Chang and Chang (2012) in this next relationship study
supported my method of data analysis.
Chang and Chuang (2012) also supported my reason for using
regression to analyze my data. In their study about attitudes on self care, it
was a relationship study where a regression model was used. The purpose of
this regression was to examine the power of basic variables, beliefs about
self-care and cues to self-care action to explain and predict self-care
behavior. Chang and Chang also employed a survey questionnaire.
Therefore, they demonstrated the flexibility of the regression analysis and its
effectiveness when employed with a survey data collection method.
Although the topic is not related to mine, I cited this study because like my
study, it used the regression to look at the closeness among these variables
where I employed a simple regression specifically with the relationship of
whether or not the predictor variables relate to the outcome variables. An
ANOVA will be used to quantify career choices.
Similar to Chang and Chuang (2012)’s relationship study using a
regression and an ANOVA, Kracher and Marble (2008), studied the
relationship between gender and morality, employing an ANOVA (Field,
2013) for the independent variable, and regression analyzed the strength of
the relationship between the independent and dependent variables, which
were gender and morality in the work place, comparing male and female
leadership traits (Chavez, 2008). In my study, my variables were STEM
courses taken by females, number of same sex role models and if they relate
to their career choices and salaries. When studying the relationship between
two variables, according to Nachmias and Nachmias (2008), regression is
employed to determine the strength of the relationship in a bivariate
analysis. This type of analysis was employed in my study, but in the form of
a simple regression.
Gaps in the Literature
There has been quite a bit of research on gender gaps when it comes
to STEM careers and the literature has given me a background or
springboard in which to begin my research. There is a wealth of background
on the reasons for these gender gaps. There were a multitude of studies
where the researchers have used a similar methodology to mine, which was
an online survey with a5-point Likert Scale. This demonstrates validity and
reliability. However, the gap that I saw in the literature that I could fill was
that there have been few studies looking at the career choices that females
make or their salaries as connected to their career choices. Few researchers
in their studies have examined that females may avoid STEM career choices
for the reasons stated in the literature such as lack of confidence,
masculinisation of STEM careers, discrimination and lack of role models
(Kenney, et al, 2012; Stout, Dasgupta, Hunsinger, & McManus, 2011).
There have been few cross-sectional studies using a survey instrument
(Harris, & Finkelstein, 2006) that have experienced the gender gap in
entering STEM careers. Also other few researchers in their studies
concentrated on the number of STEM courses taken and correlating with
making STEM career choices.
However, few researchers have actually examined the relationship
between math and science classes taken in high school and college and
career choices made and salaries earned, particularly using a regression
(Field, 2013). Furthermore, there have not been any such researchers whose
place constraints as taking place in Long Island at the four chosen
universities for this study (Nachmias & Nachmias, 2008). In addition, few
researchers in their studies used a stratified random sample, from alumni
associations like my particular study. The sample chosen were specifically
targeting females from these alumni associations that were born after 1980,
and have approximately 6 years of work experience or were in the sixth or
seventh life stage of Erikson’s life stages of development (Erikson, 1980).
Therefore, these were some of the gaps in the literature that this study can
potentially fill and add to the body of literature on females and STEM
courses and careers. This study also had objectives that were not the focus in
other studies.
An objective of my study was to determine if there is a relationship
between these STEM classes females take, their role models, and the career
choices which correlate with a certain salary level. For example, having
chosen a STEM career as a result of taking more math and science classes,
has tended to correlate with a higher paying salary. Demonstrating this
relationship was one objective of my study, this could fill in this gap in the
literature. The research method that I have chosen to use which is an online
survey has been successfully employed in this field, which was promising
for my study and the validity and reliability of the findings, once the data are
collected and analyzed.
There were several researchers that used an online survey method
which was my data collection method for this study. Alshare and Miller
(2009) studied sex traits of both males and females and employed a survey
with a 5-point Likert Scale to collect the data, which is the same method as
my study. According to Alshare and Miller, males were seen as
authoritative, individualist, and assertive. Females were seen as submissive,
and collective, meaning concerned for society and lack confidence in math
and science. This may rationalize why females may not choose STEM
careers as postulated by Carrell, et al (2010), Kenney, et al (2010) and
London, et al (2011). Since this method was valid and reliable for this study,
this survey method will work well for my study because it is a relationship
study examining a relationship between the number of STEM courses taken
in high school and postsecondary education by my sample of females from
the four chosen universities in Long Island and their career choices.
Sometimes low response rates can be an issue with any survey (Shao, 2002).
Consequently, according to
McCullough (2011), when distributing surveys, it was best not to use money
to increase response rates for an academic survey as it may induce cheating
and bias, reducing the validity of the study.
Sheaffer, Bogler, and Sarfaty (2011), also employed the same method
as my study, which is the online survey using the 5-point Likert Scale. They
tested whether or not masculinity affected and predicted how prepared one
was for preparing for an emergency situation and found that masculine traits
like assertiveness and authority helped preparation in emergency situations
more than passive traits exhibited by females. Using the survey, this may
have also strengthened the argument on why some females avoid STEM
careers because of their masculinity. Again, this helped to rationalize the use
of a survey method for this kind of study. For my study, where I examined
the relationship between STEM courses females take, their role models, and
career choices and salaries, this method is the best form of data collection
(Shao, 2002). Some advantages of online surveys included less keypunch
errors, cheaper, more global reach, greatly reduced interviewer bias, greater
interviewer control over randomization, allowed for customization by the
researcher, executive skip patterns, and logic checks
(McCullough, 2011).
Quantitative Survey and Different Methods
The survey instrument that I employed for my study was an online
survey with its many advantages such as being inexpensive, global, easy to
administer with automatic skip patterns based on responses (Shao, 2002;
McCullough, 2011). As an adjunct marketing instructor, who also teaches
marketing research, I have the skills and background to design my own
questions, which I accomplished here. I used the research questions to create
the questions to ask. My questionnaire was a 5-point Likert type scale and I
asked 26 important questions with 4 demographic questions. The survey
took 10 to 15 minutes to complete and is a quantitative closed-ended
questionnaire for a crosssectional relationship study. In the survey, I focused
on the research questions where I asked about the number of STEM classes
taken in high school and postsecondary education and their career choices in
employment which is a management function of career planning. Then I
asked about the number of role models and their career choices. To answer
the next research question, I also have questions where I asked about
salaries, using a regression. Then through an analysis of variance (ANOVA),
I analyzed across categories the relationship between the number of STEM
classes and role models individually with career choices (Field, 2013; Green
& Salkind, 2011). Then I employed a regression where I analyzed the
relationships of the independent variables of the number of STEM classes
and role models separately with their salaries, as the dependent variable, in
Research Question 3. The questions about salaries were asked as one of the
five demographic questions. When I analyzed the responses, I concluded
characteristics about the relationship between the number of STEM classes,
and role models and their relationships with the respondents’ career choices
and salaries. This is why the study was needed which was all part of career
planning, an integral part of general and employment management and
gender gaps in the scientific work place.
Summary of Chapter 2
In this chapter, I summarized and introduced a comprehensive review
of prior research conducted on this topic. Moreover, I discussed the title
check and research conducted to construct this literature review. The history
of the topic was briefly analyzed offering reasons for the gender gap in
STEM careers and why females may avoid such careers. Prior studies were
analyzed so that I could determine the gaps and where my study could fill
those gaps. Then prior research methods were analyzed in order to
rationalize my use of a cross-sectional study using an online survey
analyzing the data with ANOVA and regression (Field, 2013).
Chapter 3: Research Method
Introduction
As I stated in Chapter 1, the principal problem addressed in this study
was that women have been discouraged from taking math and science
classes during high school and college, resulting in less access to these high-
paying careers and creating a gender gap, impacting career management
(Brown et al., 2011; Correll, 2004; Hensvik, 2014; Milgram, 2011; Seibert et
al., 2013). Management consists of planning, organizing, and controlling,
and career management includes these same elements (Argyris, 1991;
Drucker, 1954; Seibert et al., 2013). Hence, this gender gap and
discouragement makes it more difficult for females to plan and organize
their careers, as well as control their financial future by having limited
access to STEM careers.
In Chapter 3, I discussed the methods of data collection and analysis
in great detail. In this study, I employ a cross-sectional, quantitative
relationship study using an online survey for data collection and ANOVA
and regression for data analysis. The relationship studied was how the
number of STEM classes and same sex role models relate to choosing a
STEM career over a non-STEM career, as well as the impact on the salary
the female receives. Variables included the number of STEM classes that
females in my sample born after 1980 have taken in high school and college,
and how many same sex role models they had to encourage them and
determine if these variables relate to their career choices and salaries. My
hypothesis was that the more STEM classes females take during their
younger years, especially in high school and postsecondary school, and the
more same sex role models they have had to encourage them, the more
likely they were to choose a high-paying STEM career. Also, since STEM
careers tended to have higher salaries, this choice should have a positive
impact on salary. This hypothesis was in accord with the literature and
particularly a claim made by Farland-Smith (2009), as well as my chapter 4
results, although the relationship of role models to career choices and
salaries was weak due to the small data set. The relationship of STEM
classes and career choices and salaries was more significant. This was
brought about by the theoretical construct of Erikson (1971, 1980), known
as Erikson’s developmental life stages. Erikson stated in his theory that what
happens in early development impacts one later in life in both the intimacy
and generativity stages of Erikson’s developmental life stages where people
are planning their careers or at the peak of their careers, in the process of
career management (Seibert et al., 2013). Therefore, how females
experienced or were exposed to science in childhood academics and
playtime can determine whether or not females are going to be interested
enough in STEM to choose a STEM career (Erikson, 1980; Farland-Smith,
2009).
The study that I conducted was a cross-sectional design using an
online survey instrument (Campbell & Stanley, 1963; Harris, & Finkelstein,
2006). Once the data were collected from the surveys, I conducted an
ANOVA to break down the career choice categories into smaller units
(Green & Salkind, 2011) and a multiple regression to assess the relationships
between the independent variables and dependent variables (Field, 2013) to
answer the research questions. For Research Question 1, the independent
variable was career choices and the dependent variable was the number of
STEM classes (Field, 2013). For Research Question 2, the independent
variable was career choices, and the dependent variable was the number of
same-sex STEM role models. For Research Question 3, the number of
STEM courses and the number of same sex role models were the
independent or predictor variables, and salaries was the dependent or
outcome variable.
Research Design
In this cross-sectional design, I employed an online survey for data
collection (Campbell & Stanley, 1963; Harris, & Finkelstein, 2006) using a
simple random stratified sample (Kalton, 1983; Rea & Parker, 2014). I used
the 5-point Likert scale since it involves measuring degrees of intensity
using intervals, making it valid and reliable (Nachmias & Nachmias, 2008).
The data analysis I used to answer the question on the relationship
between the variables was a regression. To quantify career choices, I
conducted an ANOVA (Field, 2013). It was also possible to conduct a chi
square for this nominal and categorical variable (Statsoft, 2011). Since two
of the research questions used one independent and one dependent variable,
a simple one-way ANOVA could have been used as it does not add
complexity to the analysis (Siegel & Castellan, 1988). Also, since there are
two dependent variables to answer different research questions, I needed to
run two different regressions since a simple regression cannot handle more
than one dependent variable. Questions 1 and 2 were analyzed using
ANOVA and Research Question 3 was analyzed using a multiple regression.
Scales measure degrees of attitudes and are common on surveys and
questionnaires (Nachmias & Nachmias, 2008; Shao, 2002). Examples of
common scales that are used in research include continuous rating scales,
line marking scale, Itemized rating scales, semantic scales, Guttman scaling,
and Likert scales, which are ordinal and interval scales (Shao, 2002). For my
research, I have chosen the Likert scale, which is one of the most commonly
used scales in the research. The scale consisted of assigning a numerical
value to intensity (or neutrality) of an attitude or an opinion or perspective
about a specific topic, in my case on attitudes about STEM classes and role
models. The Likert scale provided an interpretation of the intensity of items
on the scale. Responses such as strongly agree, somewhat agree, neither,
somewhat disagree, and strongly disagree are examples of responses that are
often found in a Likert scale commonly employed in surveys and
questionnaires (Nachmias, & Nachmias, 2008; Shao, 2002).
In this study, I employed a quantitative research design using an
internet survey instrument with a 5-point Likert scale. I created the
instrument, but I extracted the idea of the 5-point Likert scale from Shao
(2002) because the Likert scale has been used before in order to increase
validity and reliability (Nachmias & Nachmias, 2008). The research design
is a relationship quantitative design that observes a relationship between the
number of STEM classes taken by the sample of females drawn from four LI
universities alumni associations and their career choices as well as salaries.
There was a small pilot study testing the survey before the research began to
modify the questions if necessary to answer my specific research question
(Teijlingen & Hundley, 2001) in order to validate the survey (Nachmias &
Nachmias, 2008; Teijlingen & Hundley, 2001; Yin, 2003). The
measurements were a 5-point Liker scale using interval scales because this
research design is best used with interval scales (Nachmias & Nachmias,
2008).
Data were collected by asking a sample of originally 487, but due to
limited university cooperation and response limitations, my sample size was
reduced to 48 female alumni from four NY area universities that are the
strata about their math and science courses, same sex role models, their
career choices, and their salaries, using a simple random stratified sample
(Kalton, 1983). The online survey employed a 5-point Likert scale asking
about STEM classes, role models, and demographics (Shao, 2002).
Statistical analysis included a simple regression see if the number of
STEM courses is a good predictor of salary (Nachmias & Nachmias, 2008).
Looking at the two predictor variables of number of STEM classes and
number of role models and how close the relationship of these predictor
variables are to career choices made were analyzed effectively with a
multiple regression in Research Question 2 (Achen, 1982; Field, 2013; Gill,
2001; Morrow, 2013). In my study, an intention was to find the relationship
between the independent variables, which are the number of math and
science classes taken in high school and postsecondary school and the
number of role models; the dependent variable is salary in Research
Question 3 for the multiple regression.
In Research Questions 1 and 2 when using the one-way ANOVA, the
independent variable was career choices and the dependent variables were
the number of STEM classes and same-sex STEM role models. An ANOVA
was employed to analyze the relationship among the variables and was a
form of a regression, using a linear function (Achen, 1982; Field, 2013).
ANOVA was a way of breaking down the total variability into smaller
categories or components and assessing if the variability due to a specific
source is statistically significantly higher than the random variability (Green
& Salkind, 2011; Iverson & Norpoth, 1987) in a cross-sectional relationship
study. For a one-way ANOVA, each individual or case must have scores on
two variables, factor, and dependent variable that divides individuals into
groups (Green & Salkind, 2011, p. 183). Moreover, the ratio of these
variances is known as the F-ratio, according to Field (2013). For the
quantifying of the career choices, there were originally seven groups of
career choices where the measure is the number of STEM courses taken by
females in my sample. Due to low responses in one group, it was necessary
to combine the groups. Therefore, by combining science and math and all
nontechnical, the groups were reduced from seven to five.
Next, I employed an ANOVA to compare the five groups of career
choices to see significant differences among the groups of categories of
career choices. Furthermore, the five categories are science, technology/IT,
engineering, math, caring professions, education, and nontechnical. The
averages were ordered and a post hoc indicated which differences are
significant (Field, 2013; Nachmias & Nachmias, 2008).
For an ANOVA, populations selected must have an equal variance,
which is called homogeneity of variance (Field, 2013; Iverson & Norpoth,
1987; Morrow, 2013). Outliers of the dependent variable should be
addressed because they can increase Type 1 or decrease Type 2 errors and
reduce generalisability of results (Morrow, 2013). The assumption of
homogeneity is like the one of sphericity, which is referred to as circularity
(Field, 2013).
A regression, like an ANOVA, can be appropriate to determine if the
number of STEM classes females took and the number of same sex role
models are valid predictors of salary (Field, 2013; Kitchens, 2003; Morrow,
2013; Nachmias & Nachmias, 2008). For the salaries, a regression was
performed to determine if the number of STEM classes a female student
takes is a valid predictor of her salary. Subsequently, I employed the factor
of salary used in the linear regression, which was divided into groups.
Furthermore, to answer Research Question 2, a multiple regression
was conducted to determine the relationship between the number of STEM
classes taken and the number of same sex role models to prepare to prepare
for a STEM career and salary, since this was a relationship study and not a
comparison between two means (Miles & Huberman, 1994; Nachmias &
Nachmias, 2008). The more STEM courses a female took in high school and
college, the more likely she would choose STEM careers, which were
careers that tended to have higher the salaries than most other fields. Then
with an ANOVA, I determined which career choices have the highest
average number of STEM courses. The averages were ordered and a post
hoc test indicated which differences were significant by having a p value of
less than .05.
There were three basic underlying assumptions for a one-way between
subjects ANOVA that I considered. The first assumption was that the
populations selected must have an equal variance, which is called
homogeneity of variance (Field, 2013; Morrow, 2013). The second
assumption of this statistical test was that the observations are independent
of each other where none of the scores are related (Green & Salkind, 2011).
The third assumption is that the population from which the sample was
extracted has a normal distribution without any skewness or kurtosis
(Kitchens, 2003; Nachmias & Nachmias, 2008). However, since an ANOVA
is such a robust test, even if it did not meet all the assumptions of normalcy
and homogeneity and if there is skewness, the test is still valid (Field, 2013).
If all else fails and the p value is inaccurate, there is a nonparametric test for
ANOVA that can be used called the Kriskal-Wallis test for comparing two
or more independent samples (Field, 2013).
There were additional post hoc hypothesis tests that must be managed
once the ANOVA has been conducted. They are the Schefflé, Bonneforri,
LSD, and the Tukey tests. These posthoc tests are implemented after the
ANOVA or factorial ANOVA to determine mean difference, significance, or
nonsignificance in the p value (Gibilisco, 2011; Green & Salkind, 2011;
Morrow, 2013). The Schefflé test is a conservative test that compares all
pairs of means. The more popular and more progressive test is the Tukey
HSD test, which also compares all the pairs of the means (Gibilisco, 2011;
Green & Salkind, 2011; Morrow, 2013). Since the LSD test had the clearest
result (see chapter 4), I executed this post-hoc. Subsequently, the effect size
is generated by determining the percentage of variance, which uses the
formula of the sum of squares between divided by the sum of squares total
(Morrow, 2013). A comparison of the variance is due to the between-groups
variability, which is the Mean Square Effect, or MSeffect, with the
withingroup variability, which is the Mean Square Error, or Mserror (Green &
Salkind, 2011; Hamburg, 1983; Kitchens 2003; Nachmias & Nachmias,
2008).
According to Field (2013), the logic of the F ratio was that it is a test
that is used if differences between group means can be expressed as a linear
model. The F ratio can test these differences. If the assumption of
homogeneity is violated, one option, according to Field (2013), is to
implement corrections via the Welch procedure, which is too complicated
for designs more elaborate than a 2 x 2 design. The best thing to do is to
bootstrap the post hoc tests, use the LSD post-hoc, or the Levene test of
equality of error variances (Field, 2013; Morrow, 2013). In my case, I used
the LSD test, but did not bootstrap due to the lack of nonparametric testing
conducted.
Rationale for the Particular Method Chosen
There are several reasons that I chose the internet survey method.
Such internet surveys are less costly, making them more economical than
face to face or phone methods (Best & Krueger, 2004; Shao, 2002).
Furthermore, only such surveys have a fast turnaround time when it comes
to data collection (Case, 2007; Miles & Huberman, 1994). According to
Campbell and Stanley (1963), it was possible to provide a quantitative
description of trends, attitudes, or opinions of a population by studying a
sample of the population. From sample results, I was generalizing about the
population, but the challenge was that I had a small sample size and data set.
However, it is important to compare the sample value with that of the
population to determine sampling and nonsampling errors (Deming, 1960;
Rea & Parker, 2014). Information was taken from a large population on a
large scale using economies of scale (Case, 2007; Best & Krueger, 2004;
Shao, 2002), but reduced to a small sample due to low response.
For data collection, I used self-administered online questionnaires
(Case, 2007), instead of mail, telephone, or face to face interviews because it
was cheaper and more global (Field, 2013; Miles & Huberman, 1994;
Nachmias & Nachmias, 2008; Shao, 2002). As for some of the questions
asked on the survey on the fixed gender roles question since childhood
(Miller, 2006; Noddings, 1986), some open-ended questions were asked
about how many math and science classes the participant took and for how
many years (Nachmias & Nachmias, 2008). However, the majority of the
survey used closed-ended, categorical questions for a categorical variable,
which rationalized the use of regression (Iverson & Norpoth, 1987; Shao,
2002). There were also questions about career choices and what careers they
are presently trained for and working at (Field, 2013; Reynolds, 2007). Since
I created my own questions, a pilot may still be necessary to ensure
reliability (Field, 2013; Yin, 2003).
This internet survey was employed using reliable scales and was
distributed through a free online Internet survey service. Due to the nature of
the variables in this study, the instrument is best used with interval scales
(Field, 2013; Nachmias & Nachmias, 2008). The survey used a valid and
reliable Likert scale using both interval and ordinal scales (Becker, 1986;
Reynolds, 2007; Shao, 2002). The Likert scale is both interval and ordinal
and permits ranking (Nachmias & Nachmias, 2008; Shao, 2002), with a
specific research design. There were four steps to consider with the Likert
scale. First I compiled the scale items. Then, I administered the scale and
questions to the chosen sample for the survey. Next, I computed the value of
the scale with the first response as 1, the second as 2, the third as 3, the
fourth as 4, and last as 5 and then summed up the values. From there, I
determined the discriminate power by taking the highest and lowest values
and determining the differences between them (Trochim, 2006d). Finally, I
selected the highest power discriminates selected and tested the reliability of
the scale as explained by Nachmias and Nachmias (2008).
There was a small pilot study testing the validity of the survey using
the Cronbach Alpha before the research begins (Teijlingen & Hundley,
2001) because I am the original designer of the instrument. The pilot helped
me determine what modifications to any of the questions were needed to
increase validity (Nachmias & Nachmias, 2008; Teijlingen, & Hundley,
2001; Yin, 2003). In this case, the comparison with males who took science,
math, and technology classes was obtained from the vast existing secondary
data so there was not any experimental comparison study (Becker, 1986;
Case, 2007; Miles & Huberman, 1994; Patton, 2002; Patton, 2009;
Reynolds, 2007; Shao, 2002; Yin, 2003).
According to Campbell & Stanley, (1963), a pilot would increase validity
and reliability.
The purpose of survey research was to make an inference or
generation about a sample or population on the respondents’ attitudes,
characteristics or perceptions so that a generalization can be made about that
population (Trochim, 2006a). Here I made an inference about females who
did not participate in as many math and science classes in school as others.
These females may make different career choices in less technical, lower
paying fields (Anderson, 2006; Gilligan, 1986; Miller, 2006; Noddings,
1986, Sharp, et al, 2008). The sample was stratified and random at the level
the chosen university’s alumni association chooses the students for the
study. This relative randomness increases validity and reliability (Field,
2013; Kalton, 1983; Nachmias & Nachmias, 2008; Reynolds, 2007; Shao,
2002; Yin, 2003).
Advantages of Internet or Online Survey
Before discussing and analyzing the advantages of the internet survey
specifically, here are some general advantages of surveys in general
(McCullough, 2011). The strengths of surveys in general are that it is
versatile method in which the same questionnaire can be easily modified for
a qualitative study as an interview guide or as a quantitative study as a mail,
phone, or email/online questionnaire (Best & Krueger, 2004; Shao, 2002)..
Survey questionnaires could use various kinds of questions including
demographics, scales, questions about attitudes, opinions, and perceptions
using various types of scales including Likert Scales (Creswell, 2014; Leedy
& Ormrod, 2005; Shao, 2002). The internet survey also is less costly and
more global than most other methods. The demographics to be asked are the
age, occupation, education, and major of the participant. The survey
response rates for online surveys are generally at a higher percentage than
those of mail surveys (Baker, Hoffman, Neslin & Novak, 2009; Skalland,
2011). The response rate is defined as the number of completed surveys
divided by the number of eligible units in the sample (Skalland, 2011).
Subsequently, with the internet or online survey, (Nachmias & Nachmias
2008) argued that with the increase in the number of people that have access
to the computer, email as such, online surveys are practical because more
than 50% households have access to computers and the internet. According
to Case (2007), there is a digital divide, where some groups have more
access than others. Furthermore, Nachmias and Nachmias (2008) postulated
that online and e-mail surveying offers several advantages. For starters,
these media offered a very rapid and quick turnaround time in the survey
process. The online or email method is also faster to conduct than telephone,
especially when dealing with very large samples. This feature of economies
of scale feature makes this method cheaper to conduct because it reduces or
eliminates the mailing and interviewer cost.
Since my study was an internet survey, here were the advantages that are
specifically for surveys in the online environment. There has been an
explosion in the number of internet surveys employed to conduct research in
the last 15 years because of their many advantages (Terhanian, & Bremer,
2012). According to Comley & Beaumont (2011), internet surveys are less
costly. In addition, they could reach a global audience on the Internet.
Internet surveys are just as valid and reliable as the non-internet surveys
because reliable and valid scales such as the common 5-point Likert scale
could be used to measure what the researcher is looking to measure
(McCullough, 2011; Terhanian, & Bremer, 2012). This type of scale can be
used to measure attitudes to find out what female students perceptions
reinforce the importance for science educators to expose them to adult
professional scientists in order for students, especially female students, to
develop a better understanding of science and the role of scientists, as
conducted in the study by Farland-Smith (2009). The survey was given
before and after the experiment where the students went to a science camp
(Farland-Smith, 2009). This Likert scale is the scale I implemented for my
internet survey instrument. Using this kind of reliable scale is another
advantage to using any survey and it can be used in an internet survey very
easily.
Other advantages of internet surveys included that they are faster, they
save time and money, and they target the niche or sample directly, and can
obtain a broader sample size (Best & Krueger, 2004). Also, there is no
postage or envelopes. One can also use a data repository of email addresses
to obtain samples (Comley & Beaumont, 2011; McCullough, 2011;
Terhanian, & Bremer, 2012). The use of this kind of survey has become very
popular due to its advantages that even the skeptical Europeans are now
using this kind of survey method in their social science research (Comley &
Beaumont, 2011; McCullough, 2011; Singleton, Royce, & Straits, 1999) and
marketing studies.
Along with these advantages of the internet survey is the ability to use
stratified sampling with this methodology which was the method I
employed, similar to Chang and Chung. Chang & Chuang, (2012) employed
a stratified sample of 193 first year students, 203 second year and 207 third
year students. The students were representative across socio-economic
statuses from high to low income. The variables were socio-demographic
and the scales used consisted of a binominal scale of yes or no as well as a
pain ranking scale divided into four groups or levels of pain. In addition a 5-
point Likert scale was used to measure the adoption of self-care behavior
and beliefs about self-care related to this condition. Furthermore, internet
surveys can manage Likert and other scales interactively (Hamel, Doré &
Méthot, 2008). Since my population is Long Island alumnae females from
four chosen universities, who have been through high school and
postsecondary education, assessing the STEM classes, career choices, and
salaries, I grouped and divided them into stratifications similar to the method
used by Chang and
Chuang (2012) for their population which is an advantage of this data
collection method. The four universities were strata from the total
population of universities on Long Island, and then my sample is stratified
into a random sample of females born after 1980 who were alumni at each of
these universities in STEM majors.
Disadvantages of Internet or Online Survey
There are only a few disadvantages to using internet surveys.
According to McCullough (2011), one major disadvantage is when money is
used as a motivator. This can reduce the quality of the responses because
respondents will rush through the survey to make additional money on a
group of surveys online. This disadvantage was evident mostly in profit
marketing and not in academic research. Since my study was academic
research, I did not use money as a motivator to increase response rates
(London, Rosenthal, & Gonzalez, 2011). Comley and Beaumont (2011) also
found other disadvantages to using surveys in general as well as specifically
online ones. One of the disadvantages also was the issue with response rates
is that these rates are based on the sample and not the population which
according to Skalland (2011) did not account for the sampling frame’s
ability to undercover the target population being studied. For this reason
Skalland (2011) advocated for a realization rate, sample frame independent
that could measure the survey’s ability to identify and survey the target
population from the four universities alumnae associations. Another
disadvantage is the digital divide, where not everyone has access to the
Internet, which can reduce validity of the sample (Best & Krueger, 2004;
Case, 2007).
Other disadvantages, according to Comley and Beaumont (2011)
included lower responses if surveys are too long, and also a lack of survey
interaction (Campbell & Stanley, 1963), which may result in higher attrition.
To overcome these issues, Comley and Beaumont recommended keeping the
survey 15 minutes or less in length, avoiding complicated or repetitive
questions and making the survey more interactive using Flash player, colour
and interaction, keeps respondents interested. To combat the length issue, I
kept my survey to no more than 20 minutes long (Comley & Beaumont,
2011; Kaczmarek, Haladzinski, Kaczmarek, Baczkowski, Ziarko, &
Dombrowski, 2012; McCullough, 2011). However, this internet survey was
distributed to the alumni association who distributed to the alumni sample.
In the Appendix, I included the copy of the survey instrument that I
employed. There are other disadvantages that can be encountered when
employing an internet survey method.
These other disadvantages of using an internet survey include
government regulations on sending direct correspondence such as email
surveys. For example in Canada, the rules are stricter than in the US (Hamel,
et al, 2008). It is also important that the survey allows for confidentiality and
anonymity through a secured server or password. Another disadvantage
could be a low response rate due to surveys being found as annoying pop-
ups or spam (Terhanian & Bremer, 2012). However, this can be rectified by
having the survey open up in a new window which is how I will rectify this
issue for the alumni sample who receive the survey (Terhanian & Bremer,
2012). Also in some cases, internet surveys may have a lower response rate
than telephone or mail surveys if the survey is considered too lengthy
(Hamel, et al, 2008; Kaczmarek, et al 2012; London, et al, 2011). However,
since this was not a marketing survey, this was an academic one, I ensured
that the survey was not cluttered by pop-ups or perceived as spam. It was
sent by email through the alumni associations at the four schools used in the
sample (Kaczmarek, et al, 2012). Therefore, in conclusion, despite some of
these disadvantages for the research question have chosen, the internet
survey has still the most advantageous data collection method to answer the
research question (Field, 2013; Shao, 2002).
Target Population and Sampling Procedures
The participants were recruited by the prospective alumni associations
after I have called and contacted each association and have informed them of
the study (Field, 2013; Kalton, 1983). I received permission and then each
association randomly selected from their alumni who were born in 1980 or
later (Field, 2013). Due to limited cooperation by the universities, the
sample size was drastically reduced, thus I made the survey available also to
the Walden pool of participants and the end total of the sample resulted in 48
due also to low response. The method was an internet survey where these
alumni association were able to provide the participants access this survey
using a secure password to increase control and email it to the alumni (Case,
2007; Shao, 2002). The alumni associations acted as the gatekeepers that
helped me gain access to the participants who were students in four schools
on Long Island, chosen for this study. The respondents were chosen at
random by the gatekeepers (Campbell & Stanley, 1963). This randomness
increased validity and reliability. This was accomplished through a sampling
distribution of the means (Morrow, 2011; Statsoft, 2011). The samples were
extracted randomly or systematically, depending on what was easier for the
alumnae associations. In both cases, each subject had an equal chance of
participating in the survey, which increases validity (Case, 2007; Shao,
2002) because the systematic starting point is random (Nachmias &
Nachmias, 2008).
For the mechanics of the survey, the questions were mostly closed
ended with few open ended questions (Shao, 2002). The participants
completed and returned the survey online anonymously through a secured
website where the participants filled out the survey through an interaction
with the website, which was set up and accessed through their alumni
association. I had access to this site as the researcher but I did not know who
the respondents are.
With sampling, Tuten (2010) cautioned against coverage errors in an
internet survey, reducing randomness. The qualifier of the alumni for this
study was that the participants must be born after 1980 and be members of
the GenY/Millennial generation, normalizing for years of experience when
figuring salaries.
In the analysis, when evaluating sampling accuracy, coverage errors
signify that there are some people who have no chance of being selected for
the study, and for example, this may be those without internet service in an
online study such as this one (Deming, 1960; Tuten, 2010). This is only a
concern if access to the internet is a major issue, resulting in a digital divide
(Case, 2007), but in Long Island, Internet accessibility is fairly universal.
However, this was not an issue in my study since the sampling frame was
random or systematic starting at a random point. In this case, sampling
frame error was more of a concern as this was an error in the sampling
frame. This kind of error was more difficult to minimize since there were no
lists of web users or email addresses, and IP addresses are unique to
machines not people. The unit of analysis to be used was individuals. In this
case, it is female students from the alumni associations of these four chosen
universities, randomly chosen by the gatekeepers.
Participants/Population
The population in which the sample was drawn was from the female
alumni of the four sampled universities in Long Island born after 1980. The
sample was randomly selected and stratified through these alumni
associations of these universities. The survey was distributed through these
associations and through the alumni associations; the respondents had access
via a password through their university, giving control to respondents (Case,
2007; Nachmias & Nachmias, 2008).
To be eligible for the study, one must be a female and must be born in
1980 or later and have received either a bachelors or masters from one of the
four universities chosen in the study. Then the participants will be randomly
chosen by the alumni association (Field, 2013). Moreover, since the sample
size was drastically reduced, I expanded it slightly by making the survey
also available in the Walden pool of participants.
Informed Consent
Participants 18 Years of Age or Older
Respondents were invited to take part in this very important study on
the impact that the number of STEM courses females take in high school
and postsecondary education and the number of female role models a female
has on their career choices and salaries. This study was sponsored by a
programme in the Business Management and Technology Department at
Walden University, under the auspices of the IRB. The intent of this study is
merely to extract your perspectives and inputs on this relationship. This form
is part of a process called “informed consent” and allows each respondent to
understand this study before deciding whether to take part. This study is
purely voluntary and causes no harm to the respondents (Nastasi, 2009).
Furthermore, confidentiality is of utmost importance.
Sampling Strategy and Defense of the Method-Sample Procedure
Randomness increased validity and reliability of the sample by
insuring equal chance of participation. However, this sample used in this
study was also a stratified sample using female students and possibly some
males born after 1980 who were students at the chosen universities and are
now alumni. Hence, this was a stratified simple random sample (SRS)
(Kalton, 1983). Therefore, bias was reduced. The rationale for using a
random sample or a systematic sample with a random starting point was that
each person has an equal chance of participation in the study, which reduced
selection bias (Case, 2007; Field, 2013; Nachmias & Nachmias, 2008,
Patton, 2009). This study had the elements of a statistical relationship
(Nachmias & Nachmias, 2008; Shao, 2002). This was a relationship study
using a survey instrument, with a random sample, using sampling
distribution of the means (Nachmias & Nachmias, 2008). The sample was
randomly drawn by each alumni association at each sample university to
randomly extract the female alumni born after 1980 for this study. Also, the
many advantages of the online survey were mentioned in the previous
section.
A regression could answer research questions like how strong is the
relationship between (Achen, 1982; Field, 2013; Morrow, 2013) the
independent and dependent variable which for example, in my study was the
number of STEM classes a female takes and the number of same sex STEM
role models she has impact the career choices she makes and salary she
earns. This regression could and did indicate the strength of the relationship
between these variables. The two predictor variables were the number of
STEM classes and same-sex STEM role models and the outcome variable is
the career choices. This analysis indicated which predictor variable has the
strongest relationship with the dependent variable, which was explained in
chapter 4.
How the Sample Was Drawn
The sample were randomly drawn or systematically drawn with a
random starting point, from the alumni associations from four universities in
Long Island, explaining my research, its purpose, and ensuring dignity and
confidentiality (Kalton, 1983) randomly extracting alumni from these
universities who were born after 1980. The stratification was that I sampled
females born after 1980 in a specific area which consists of four schools on
Long Island, being taken from a subgroup, along with the few participants
from the Walden pool (Statsoft, 2011). It was slightly difficult to obtain the
sampling frame of the alumni emails due to confidentiality issues (Kalton,
1983). Therefore, I had to contact the alumni associations and ask if they can
distribute the internet survey to the respondents. In addition, the issue of
some females refusing to participate can be an issue. To avoid selection bias,
a strict probability mechanism must be used (Kalton, 1983).
Also, another way to avoid selection bias was to give each sampling
frame element a known and nonzero probability of selection avoiding
missing elements (Field, 2013; Kalton, 1983; Morrow, 2011; Shao, 2002).
These missing elements was a slight limitation and a weakness to the study,
mitigated with additional, specific sampling frame lists of these female
employees at each of the companies in the sample (Kalton, 1983).
Since this was an internet survey, the missing responses may be random
from females who do not respond to the alumni association (Case, 2007).
The process to obtain the sampling frame from each of the alumni
from the four university’s alumni centre was through a lottery in choosing
which female alumnae were studied of which the alumni association acts as
gatekeeper (Kalton, 1983). The stratification was geographic taking place in
Long Island. An email letter was sent to the alumni association of these four
universities discussing the purpose of the study including a privacy and
confidentiality clause (Kalton, 1983; NIH 2008). Then a follow-up phone
call was made to the alumni directors of each of the sampled universities.
Before anyone was contacted, IRB approval was obtained since this study
involved human subjects (NIH, 2008). To perform the lottery method, the
alumni association forwarded the survey link to the alumnae randomly. I did
not have any contact with the respondents directly, ensuring anonymity.
A weakness of the sampling frame was the possibility of missing
elements where it made it possible that a particular female alumna had no
chance of participating in the study (Field, 2013; Kalton, 1983). This failed
to represent the entire population. To mitigate this situation, I tried to obtain
several lists from each alumni association to make sure each female alumni
participant has an equal chance of inclusion. However, I was unable to do
so; thus, I had to trust the alumni associations to be the gatekeepers (Kalton,
1983).
Sample Size
The frames for the sample were the female alumni at the four schools
studied, to whom the alumni association then forwarded the survey (Kalton,
1983). This is the population being targeted to answer the research questions
about the relationship between the number of STEM classes and role models
and how these relate to career choices and salaries. This sample size was
chosen since originally it was a large enough sample to obtain external
validity and generalization to the experiences of the total population
(Burkholder, 2010; Field, 2013; Nachmias & Nachmias, 2008). The original
chosen sample size was 487 from strata of four universities in the NY area
or the number of universities who cooperate with my study (Kalton, 1983).
However, due to low response, the sample data set was drastically reduced
to 48. For this reason, some respondents from the Walden pool of
participants were permitted to participate. To calculate the G power analysis,
the effect size is medium at .50, which is acceptable for Cohen’s D as
postulated by Field (2013) and Sheperis (2014). Furthermore, the alpha or
significance level is .05 and these calculations are for a two-tailed test. This
alpha means that 5% chance of being wrong when the null hypothesis was
rejected, about the relationship between the independent and dependent
variables. The error problem is .05, and 1 – b is .80. N = 487 which is my
total intended original sample population for all four universities or the total
number of cooperative universities. However, low response decreased it
substantially to 48. Power is one – beta and this helps to avoid type II errors
where a researcher fails to reject the null hypothesis (Sheperis, 2014).
Furthermore, the response rate was defined as the completed surveys divided
by the eligible surveys in the sample (Skalland, 2011).
This is written as:
Complete Interviews/
(Observed Eligibles) + e_(Units with Eligibility
Undetermined); E represented the assumed rate of eligibility among the
units for which eligibility status has not yet been determined. Some
limitations included the inability to account for coverage, meaning those
without internet access, or the sensitivity to the assumed value of e
(Skalland, 2011). To remedy this issue, this solution for e formula can be
used:
e = Observed Eligibles/
(Observed Eligibles) + (Observed Ineligibles)
Response rates could vary depending on the survey, its length, and the
method being used. These response rates can vary from 35% or 36,8% to
50% because of the limitations listed above (McAllister, 2015; Skalland,
2011). To err on the side of caution, I used 35M or 36,8%. Using 356 as the
original population from my G Power, and multiplying 36,8% response rate,
my total population becomes 487, which are approximately 122 per school.
36,8% of 356 is 131. The sample who responded was 48. These figures were
rounded to the next number.
This sample must be restricted using proportion stratification (Kalton,
1983). This is a single-stage sample, stratified by geography (Kalton, 1983).
One of the factors used to determine the sample size is confidence interval
which in this case is 95%. A 95% confidence interval is where there is a
95% likelihood that the interval contains the true but unknown parameter, or
population value within the range, in this case the mean difference
(Gibilisco, 2011; Green & Salkind, 2011, Kalton, 1983; Kitchens, 2003;
Morrow, 2011). According to McAllister (2015), confidence is the inverse
of significance or 1 – α. The level of significance which is .05 was the
probability of a type one error and signified the acceptable risk that I was
willing to accept in the event that the ANOVA or regression reveals an
effect that may not exist (McAllister, 2015).
Another factor used in determining the sample size was the possibility
of nonresponse (Kalton, 1983; Shao, 2002). The sample size was determined
by the forecasted percentage response rate expected, which is 35% and the
confidence interval, which is 95% (Field, 2013; Kalton, 1983; McAllister,
2015; Statsoft, 2011). The same confidence interval and p value were
applicable to the statistical tests used in my study.
The other way that the sample size was determined was by the alpha
which is usually .01 or 05. The larger p or significance value resulted in a
larger region of rejection for the hypothesis which is that participation in
math and science may increases career success for females through choosing
higher paying STEM careers (Burkolder, 2010; Nachmias & Nachmias,
2008).
The Research Questions and Hypotheses
Research Question and Hypothesis 1
Research Question 1: What is the relationship between the number of
STEM courses taken in high school and postsecondary school by females
and their career choices?
Hypothesis One
Ho: The means of the number of STEM classes are the same for different
career choice categories
H1: At least one of the means of the number of STEM classes is not the
same for the different career choice categories Hypothesis in
Statistical Terms
Ho: 1=µ2=µ3=µ4=µ5= 6= 7
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different 1≠µ2≠µ3≠µ4≠µ5≠ 6≠ 7
1 to 7 are the factor groups. The independent variable is the career choice
categories.
The original 7 factor groups are defined statistically below.
1= career choices for females in science
2= career choices for females in technology/IT
3= career choices for females in engineering
4= career choices for females in math
5= career choices for females in caring professions
6= career choices for females in education
7= career choices for females in nontechnical fields like legal, business,
etc. These categories were reduced to five as follows:
1= career choices for females in science and math
2= career choices for females in technology/IT
3= career choices for females in engineering
4= career choices for females in nontechnical (soft sciences like
business, poli sci, legal etc)
5= career choices for females in caring professions
In this statistical construct using an ANOVA, in these factor groups,
the dependent variable was the number of STEM classes and the
independent variable is the career choices. Using a one way ANOVA, I
determined if the average numbers of STEM classes taken were different
across factor groups which are the five career choice categories. I intended
to demonstrate that females who choose STEM career categories tend to
take more than STEM classes than those who do not choose such career
categories. My intention was to retrospectively demonstrate that the number
of STEM courses taken, are different by career choice categories which were
the factors. In other words, I wanted to test if career choice categories were
related to the number of STEM course taken in the past. I could then also do
a set of multiple comparisons to see if some of the categories are the same
statistically.
I compared these four STEM groups of science/math, technology/IT,
engineering, versus three non-STEM groups of caring professions,
education, and nontechnical. Photonics and research and development were
included in engineering, as were electrical, mechanical, civil, and aerospace
engineering. Caring professions were healthcare, nursing, medical, and
home health aides. Education included teachers, professors, or anyone who
works in a school district or postsecondary institution. Nontechnical includes
those professions that are not in a STEM, caring, or educational profession
(including business, administrative, service, retail, manufacturing, and
legal).
In using ANOVA which was a procedure to test the hypothesis in order to
evaluate the differences in the means among the seven groups below
(Iverson & Norpoth, 1987; Morrow, n.d.), I am investigating the differences
between these groups of career choices for both questions one and two. I
found it necessary to reduce the categories due to low responses in certain
categories. For example, since there were no respondents who chose
engineering, I eliminated this category to reduce the categories. I also
combined math and science and reduced the categories from seven to five.
The hypothesis was that there was a positive relationship between the
number of STEM classes taken in high school and postsecondary school and
choosing a STEM career.
I employed the post-hoc test to identify which courses have the
highest significance. This was important because ANOVA did not tell which
of the categories were different, only that at least two of the categories are
different. Since Ho: was rejected, there was statistical support that the
impact or relationship between the kind of career choice chosen based on the
number of same sex role models or number of STEM courses taken and
salaries received is different. The post hoc test determined the greatest
differences (Field, 2013). Under the null hypothesis, using the ANOVA, the
relationships were equal across factor groups. However, since this was not
the case, the null hypothesis was rejected. One of the inherent issues with
any cross-sectional relationship design is researcher bias, which I tried to
control to the best of my ability.
Research Question and Hypothesis 2
Research Question 2: What is the relationship between the number of
STEM role models in high school and postsecondary school and their career
choices?
Hypothesis Two
Ho: The number of female STEM role models in high school and
postsecondary school are the same for the different career choice categories
H1: The number of STEM role models in high school and postsecondary
school are not the same for the different career choice categories
Ho: 1=µ2=µ3=µ4=µ5= 6= 7
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different 1≠ µ2≠µ3≠µ4≠µ5≠ 6≠ 7
The factor groups used in Research Question 1 as well as the statistical
analysis will be the same for Research Question 2 as was indicated in
Research Question 1. The only difference is that the dependent variable is
the number of female STEM role models instead of the number of STEM
classes taken.
In this statistical construct using ANOVA, in these same factor
groups, the dependent variable was the number of same-sex STEM role
models and the independent variable is the career choice categories. Using a
one way ANOVA, I determined if the average numbers of STEM same-sex
role models were different across factor groups which are career choice
categories. I tried to demonstrate that those females who choose STEM
careers tend to have more STEM same-sex role models than those who do
not have such role models. The hypothesis was that there is a positive
relationship between the numbers of STEM classes taken in high school and
postsecondary school and choosing a STEM career (Farland-Smith, 2009).
Under the null hypothesis, using an ANOVA, the relationships are equal
across factor groups. The more same-sex STEM role models a female has,
the more likely she is to choose a STEM career. The factor groups were the
same for both questions one and two. Again, similar to Research Question 1,
my intention is to retrospectively investigate the number of STEM same-sex
role models are different by career choice categories which are the factors.
In other words, I wanted to test if career choice categories were related to
the number of same-sex STEM role models females had in the past. I could
then also do a set of multiple comparisons to see if some of the categories
are the same statistically.
Research Question and Hypothesis 3
Research Question 3: What is the relationship between salaries and the
number of STEM courses taken in high school and postsecondary school by
females, and the number of same sex role models?
Hypothesis Three
Ho: The salaries are independent of number of STEM courses in high school
and postsecondary school and/or role models.
H1: Salaries are dependent on the number of STEM courses in high school
and postsecondary school and/or role models.
H0: 1= 2=0, both betas are zero
H1: at least one s not equal 0
Y= o+ 1X1+ 2X2 where X1 is the number of STEM courses and X2 is the
number of same sex role models and Y=salaries
The number of STEM courses and the number of same sex role
models were the independent or predictor variables and salaries is the
dependent or outcome variable. This was a multiple regression. In this
research question, I attempted to establish a linear relationship between the
independent and dependent variables (Field, 2013). The hypothesis was that
the number of STEM classes taken and the number of same-sex STEM role
models have a positive relationship with a higher salary since if these
conditions exist, it is more likely females will choose STEM careers which
tend to have higher salaries. With an .05 alpha means I had only 5% chance
of being wrong when H0 is true and is rejected, about the relationship
between the independent and dependent variables about the relationship
between participation in math and science classes through the number of
classes females took, the number of female role models and their impact on
career choices for females facing barriers to entering these STEM fields.
Thus, the confidence interval was 95% which means there was a 95%
likelihood that the interval contains the true limits where the population
mean is likely to fall (McAllister, 2015). The confidence interval was the
known parameter of participation in math and science correlate with
choosing a STEM career success with the ability to earn a higher salary than
without choosing a STEM career (Burkolder, 2010; Field, 2013; Gibilisco,
2011; Green & Salkind, 2011; Nachmias & Nachmias, 2008). It was the
proportion of variance in my dependent variable of salaries that was
accounted for by my set of independent variables. This was the overall effect
size for the regression I used for my study (Field, 2013; Morrow, 2013).
The last factor was the effect size of the sample size, determined by
how strong the relationship was between participation in math and science
and career choice of a STEM career, which translates to higher pay and
success (Burkolder, 2010; Gilligan,
1986; Nachmias & Nachmias, 2008). The effect size was the mean
difference divided by the standard deviation. Since the original sample size
was fairly large, this should help the strength of the relationship between
these two variables. However since due to low response, the sample was
dramatically reduced, this contributed to a much weaker relationship among
the variables of role models versus the other variables. The p value or
significance means that the treatment had an effect and therefore, and would
confirm it through a significance test which rejects Ho (Field, 2013; Green
& Salkind, 2011; Nachmias & Nachmias, 2008; Statsoft, 2011). If p was less
than ,05 or ,01 the relationship between these two variables is significant
(Nachmias & Nachmias, 2008). This means there was a 5% of obtaining the
data obtained if no effect exists, then I must determine if I am confident
enough to accept the effect on the sample is genuine, meaning that the more
STEM courses taken by females in the sample, the more likely the females
would choose a STEM career (Field, 2013). The results are in Chapter 4. At
each university originally I proposed the sample of 487/4 = 122 from each
university alumni association will be haphazardly selected as a simple SRS
(Field, 2013; Green & Salkind, 2011; Kalton, 1983; Nachmias &Nachmias,
2008). However, the low response rate resulted in a sample of 48 with a
small data set. Thus, to mitigate this issue, some respondents from the
Walden pool of participants were permitted to participate. They are included
in the 48 sample size.
My sample should have been 487 people from four Long Island
universities, 122 at each university, if I would have received100%
cooperation from the universities. To reduce bias, it is necessary to be
mindful of outliers and residuals and minimize nonresponses (Morrow,
2011). The predictor variable of courses taken and the variance of the
residual terms, which were normally distributed, must be constant. My
sample size was above 10, it was 48 after the reduction, which was large
enough and satisfactory for the regression. According to Burkholder (2010)
and Trochim (2006d), one must consider the effect size, and alpha in the
calculation of sample size (Buchner, Faul, & Erdfelder, (n.d.).
To compute my sample size to achieve 80%, use the alpha α and 1 –
β, which is normally .05 and .8 respectively. At 80%, the power is .20
(McAllister, 2015). The effect size is eta squared and is calculated by
dividing the effect of interest by the total amount of variance in the data
(Field, 2013). Then the amount of participants that is needed to detect the
effect is calculated (Field, 2013; Kalton, 1983). The sample size effects
significance. In a small sample large differences can be non-significant.,
which was what occurred when the sample size was reduced. The sample
size required depended on the kind of effect that I tried to detect meaning
how strong was the relationship being measured and how much power
needed to detect these effects (Field, 2013; Kalton, 1983; McAllister, 2015).
Power is the probability that a significant difference is detected among
groups after conducting a test such as an ANOVA, or regression
(McAllister, 2015). 1 - β = power. Usually the larger the sample size, the
better. The small sample size that resulted was a limitation of my study. The
sample size required depends on the effect. When I used a regression with
the dependent variable as salaries and the independent variables as the
number of STEM classes taken by the sampled females and the number of
same-sex STEM role models, then the R squared was the squared multiple
relationship. R squared is the multiple correlation squared, which is the
overall effect
size. It is the proportion of variance in the dependent variable of the female
salaries, accounted for by the independent variables of the numbers of
STEM classes and same sex role STEM models for Research Question 3
(Field, 2013).
G Power Calculation
According to Buchner, & Erdfelder, (n.d.) The G power is the analysis
that I used to determine the original sample size for my study. The power is
the odds that one can observe a treatment effect when it occurs or the odds
of saying that there is a relationship, difference, gain, when in fact there is
one such as the relationship between the number of STEM classes and
female role models has on career choices (Trochim, 2006d). A lower alpha
makes a type one error less likely. In contrast to a priori power analyses,
post hoc power analyses often make sense after a study has already been
conducted (Faul, Erdfelder, Buchner, & Lang, 2009). Power includes sample
size, B, effect size, and C, alpha level. A post hoc analysis is computed as a
function of the population effect size parameter, and the sample size used in
a study (Faul, et al., 2009). The level is the chance of error that researchers
are willing to take in determining statistical significance. An alpha level
at .05, means the willingness to accept a five percent chance of error in their
statistical analysis, if the HO is correct and rejected, which goes along with a
confidence interval of 95% (Field, 2013; Morrow, 2011; Sheperis, 2014). If
one used .10, the chance of finding significance increased. Power should be
set at .80 which means that as a researcher I have an 80% chance of finding
a significant difference between my variables, and only 20% chance of
committing a type II error (McAllister, 2015). The effect size was the impact
of any treatment or intervention or the impact of the number of
STEM classes and role models on career choices.
There were a few different types of effect size analyses. One measures
mean differences and the other measures proportion of variance (Field,
2013; Sheperis, 2014). When it came to measures of association, the most
common effect size calculations were Eta-squared, R-squared, Omega-
squared, and the Phi coefficients.
Eta squared is simply the sum of squares between, divided by the sum
of squares total. If the eta squared is weak then that means that the number
of STEM classes and role models have little effect on career choices or
salaries. R-squared is the proportion of variance that was explained when
examining the association between variables, which ranges from zero to one
like the number of STEM classes, and role models and how they relate to
salaries using a multiple regression. The phi coefficient is the standard effect
size calculation for a Chi Square used for a nominal variable (Sheperis,
2014; Stephens, 2004). “Career choices” was a categorical variable which I
quantified, as mentioned earlier in the ANOVA discussion. Phi was related
to the correlation or relationship and it estimated the extent of the
relationship between two variables, such as number of STEM classes
females took and career choices. When a researcher calculates omega
squared as the effect size, it is the sum of squares between, minus the
number of groups such as the number of females who took three or more
STEM classes each year since high school.
With regard to measures of difference, the two most common
calculations were Cohen’s D and Cohen’s F. Cohen’s D simply took the
mean of group one, minus the mean of group two, and divide by the error
term as an example. Cohen’s D is used with t tests and ANOVAs. Using,
post hoc analyses, calculate Cohen’s D to determine the effect size for each
pairwise comparison (Field, 2013; McAllister, 2015; Morrow, 2011;
Sheperis, 2014). I employed an ANOVA and regression with a post hoc test.
According to Cohen, when interpreting F .10 is a small effect size, .25 is a
medium effect size, and .40 or larger is considered a large effect size,
meaning that for example .10 means that the number of STEM classes and
role models have a small effect on career choices, .25 means a medium
effect and .40 is a large effect on career choices.
To calculate the G power analysis, the effect size is medium at .50,
which is acceptable for Cohen’s D as postulated by Field (2013) and
Sheperis (2014). Furthermore, the alpha or significance level is .05 and these
calculations are for a twotailed test. The error problem is .05, and 1 – b
is .80. N = 487 which was my total original sample population for all four
universities or the number of cooperative universities, which due to low
response rate was reduced to 48. Power is one – beta and this helps to avoid
type II errors when a researcher fails to reject the null hypothesis (Sheperis,
2014). To prevent these errors, this calculation employs the output non-
centrality parameter represents the degree to which the null hypothesis is
false, so that type II errors can be prevented (Quinn, 2014). The output
parameters also provided critical value, degrees of freedom, and the sample
size for each group of sampled participants, the total sample size, and the
actual power for my study.
Here the how my sample size was estimated. This was a priori power
analysis which computes the required sample size, given alpha level, power,
and effect size as indicated. My effect size turned out to be .15 which was
low to medium. My power 1 – b was set at .80 and my alpha is .05. I
conducted a regression (Achen, 1982) fixed model, single regression
coefficient. I also conducted a priori which computed sample size, effect
size, power, and alpha. My non-centrality is 3.693 my critical T is 1.9789,
degree of freedom is 86 and my sample size is 122. I believe that I must
survey 122 alumni randomly at each of the four universities in my sample.
My actual power is .95 in the output, but was set at ,80 and my partial r
squared is .5 with a residual variance of 1.
Confidentiality
It was difficult to obtain the sampling frame of alumni due to
confidentiality issues (Kalton, 1983), which must be adhered to in order to
protect the privacy and dignity of respondents. Two phase sampling was
difficult because of confidentiality; the alumni associations were not willing
to provide a list of students for this sample (Kalton, 1983). To mitigate this
situation, I assured the alumni associations, who were the gatekeepers
(Campbell & Stanley, 1963), that the IRB of the university insured
confidentiality, dignity, and anonymity for all participants (Nastasi,
2009).This study was purely voluntary and all answers are confidential. I
protected the rights, dignity and confidentiality of human subjects. Their
identities and responses were held in the strictest of confidence. The alumni
associations agreed to post the survey on websites or newsletters for students
to respond.
Procedure
The respondents were asked some classifying questions.
The respondents were asked about their own experiences in STEM
classes, and careers, and with female role models in these fields.
The respondents were asked to quantify her classes, role models, and
rate her experiences in STEM and the impact on her career choices.
The respondents were asked some demographic questions such as
career choice, salaries, and income.
For the respondents who agreed to participate, the survey took
approximately 25 minutes to complete and there was no compensation
for taking part in this study.
This study was purely voluntary whether or not one chose to be in the
study. No one at Walden University or current organization would treat the
respondent differently if she decides not to be in the study. If a respondent
decided to join the study now, the respondent was still free to change her
mind later and opt-out.
Risks and Benefits of Being in the Study
Participating in this study did not pose any risk to safety or wellbeing.
The benefits of the study were that the responses would help add to the body
of knowledge as to why females may not choose STEM careers and have
role models. This study also can help encourage more females to choose
STEM classes and careers by encouraging females in these fields and
making these fields more fun and interesting to females.
Any information provided by a respondent was kept confidential. In
any type of report that might be published, I will not use any personal
information for any purposes outside of this research project. Research
records are kept in a secure file and only I have access to the records.
Additionally, data will be kept for a period of at least 5 years, as required by
the university.
Geographic Location
The study took place online and the survey was conducted through the
internet.
The population participants or sample being studied are from four
universities on Long Island. They did not have to live in Long Island, as
long as they are alumnae of one of the four chosen schools on Long Island in
this study. Online surveys have higher response rates than postage mail
surveys (Shao, 2002; Yin, 2003). Moreover, online surveys are less costly to
administer.
Instrumentation and Materials
The scale used to measure participation in math and science and
career choices and success is the Likert Scale (Belch & Belch, 2004; Shao,
2003). The levels of measurement that were important for this research are
the ordinal and interval scales, using a Likert Scale. Likert scales are
flexible and they measure the intensity of attitudes and emotions in a
variety of applications (Nachmias & Nachmias, 2008; Shao, 2002).
Nominal scales will be used for demographic information because they are
mutually exhaustive and exclusive. An example is using one for male and
two for male (Nachmias & Nachmias, 2008; Newman, Ridenour, Newman,
Mario, & DeMarco, 2003; Shao,
2002). The information that was useful to my study was the gender, the
respondents’ occupation, their salary, and the math and science classes
taken in high school and postsecondary school. The instrument used was an
internet survey questionnaire which I created and structured in the shape of
a funnel with the easy questions to start, the tough questions in the middle
and the demographics at the end (Shao, 2002).
For the structure of the questionnaire, the easy questions were in the
beginning. The principle questions on the actual discussion of the
perceptions on the social development with the connection between
participation in math and science classes, female role models and their
impact on career choices which may also impact on career success, and
salary, were in the middle of the survey questionnaire, using the funnel
sequence. The demographic questions were at the end (Shao, 2002). Also,
the biggest problem that I made sure I was careful of was to reduce bias
when I asked the questions. Since I have passion about this topic, being a
professional woman who has experienced discrimination, I had to be careful
to word the questions objectively as to not bias the results (Belch & Belch,
2004; Case, 2007; Shao, 2002; Yin, 2003). Bias lowers validity and
reliability of the survey questionnaire and the results. Content Validity
assured that the content is authentic, authoritative and the scales measure
what they are supposed to. Therefore, it is important to make sure that the
scale properly measured the variables of encouragement and fixed gender
roles (Miles & Huberman, 1994; Nachmias & Nachmias, 2008). A pre-
tested pilot study was conducted to increase validity and reliability of my
survey instrument.
The instrument was pre-tested using a pilot study in order to increase
validity and reliability by conducting a Cronbach Alpha (Campbell &
Stanley, 1963; Yin, 2003). Since I used my own questions to answer the
research questions, the pilot became necessary in case I had to modify any
questions. Prior knowledge of the instrument and testing could also bias a
study and be a threat to internal validity and external validity or
generalisability (Becker, 1986; Case, 2007; Shadish, Cook & Campbell,
2002; Shao, 2002; Yin, 2003). Other threats to internal validity included the
personal experience of the participant, history, growth, and maturation from
a study (Becker, 1986; Case, 2007; Shadish, Cook & Campbell, 2002; Shao,
2002; Yin, 2003).
Relation of Survey Questions to the Research Questions
I used the research questions to create the survey questions to ask. The
first research question asked what kind of relationship is there between the
number of STEM classes females took and career choice categories. Here I
ask questions related to both STEM classes and career choices with
categories. In the second question, I asked what kind of relationship was
there between the number of STEM classes and salaries, and I asked about
salaries as a demographic question, using a Likert scale and demographic
questions. For answering these questions I used a Likert scale and
demographic questions. Questions 1 and 2 both use an analysis of variance
(ANOVA) and questions 3 uses a multiple regression with two independent
variables.
My questionnaire used a5-point Likert type scale and I asked 26
important questions with 4 demographic questions. The survey took 15 to
20 minutes to complete and is a quantitative closed-ended questionnaire for
a cross-sectional relationship study. The questions focused on the research
questions which ask about the number of STEM classes that females took in
high school and postsecondary education and their career choices in
employment which is a management function of career planning. Then the
questions also ask about the number of role models and their career choices.
To answer the next research question, I also had questions that ask about
salaries. Then through an ANOVA, I analyzed across categories the
relationship between the number of STEM classes and role models
individually with career choices (Field, 2013; Green & Salkind, 2011).
Then I employed a multiple regression which analyzed the relationships of
the independent variables of the number of STEM classes and role models
separately with their salaries as the dependent variable. The questions about
salaries were asked as one of the five demographic questions.
Establish Reliability of the Instrument
There are two types of reliability, which measured consistency over
time through pre and post testing. Reliability determines if our errors are
systemic or random (Yin, 2003). To counteract the reliability issue of
limitations of pre and post testing, the parallel forms technique could be
used (Campbell & Stanley, 1963). This method was conducted by creating
two parallel versions of the survey instrument, and administering both
versions to the two groups, being sampled. Then the two sets of measures
are correlated to increase reliability. Since I conducted a pilot, this parallel
form was necessary (Sherman, 2004; Yin, 2003). Split-half method of the
Cronbach Alpha estimates reliability by treating each part of an instrument
as a separate scale, to increase generalization. Each part of the 5-point Likert
subscales were treated as separate instruments when figuring out the
Cronbach Alpha which measures reliability and should measure around .7 or
.8. This increased external validity (Nachmias & Nachmias, 2008), since I
created the instrument.
The variables were measured on Likert Scale, which is both ordinal
and interval
(London, Rosenthal, & Gonzalez, 2011; Nachmias & Nachmias, 2008; Shao,
2002). The
scale uses the one to five ranking scales that could be used for the
independent variable. Also a nominal scale is advantageous for the
demographic questions on job categories for career choices and lists of math
and science classes taken by females in the sample and their career choices
the latter of which was a categorical variable.
Also, if the instrument is pre-tested using a pilot study, this increased
reliability, which is consistency over time (Yin, 2003). If the results can be
generalized, they can be replicated over time. Increased validity increased
reliability (Yin, 2003). They were directly related (Nachmias, & Nachmias,
2008; Reynolds, 2007). In order to increase reliability, I also conducted a
pilot before the general survey (Teijlingen & Hundley (2001), using the
same instrument, to increase reliability (Shao, 2002; Yin, 2003). This is
important because if a test shows consistent results, that means the
instrument is credible and consistent making the results useful for the survey
(Shao, 2003; Yin, 2003).
Establish Validity of the Instrument
Content Validity assured that the content is authentic, authoritative
and the scales measure what they were supposed to, which is strength. Using
a broad literature on social development, STEM classes taken, their career
choices, affect on salaries and gender (London, Rosenthal, & Gonzalez,
2011) has increased the validity of the content. Therefore, this literature will
be used as a springboard, to create the questions on the scale to ensure
content validity (Nachmias & Nachmias, 2008). The items of content
measured were the ones that I intended to measure. Content validity
measured all the attributes that are intended to measure (Field, 2013;
Gibilisco, 2011; Nachmias & Nachmias, 2008; Statsoft, 2011). Therefore, it
was important to make sure that the scale properly measures the variables of
participation in math and science courses and access to professional higher-
paying careers, for example (Miles & Huberman, 1994; Nachmias
& Nachmias, 2008). Under content validity, there is face validity. This is the
subjects’ evaluation of the investigation and the appropriation of the
instrument. Sampling validity is when the population or total number of
cases is sampled adequately for the instrument (Gibilisco, 2011; Miles &
Huberman, 1994; Nachmias & Nachmias, 2008; Shao, 2002; Yin, 2003). By
using a random sample this ensures sampling validity and I used a simple
stratified random sample (Kalton, 1983).
Construct validity is when the hypothesis or construct is measured as
intended. The hypothesis is measured using the p value to determine
significance of the relationship between the two variables (Nachmias &
Nachmias, 2008). This kind of validity gives meaning in a descriptive sense
to the instrument (Nachmias & Nachmias, 2008). To ensure this construct
validity, I ensured the 5-point Likert Scale used employ questions that
coincide with the hypothesis about participation in math and science, and
career choice which correlate with increased salary (Field, 2013; Gibilisco,
2011; Gilligan, 1986; London, Rosenthal, & Gonzalez, 2011; Noddings,
1986; Miler, 2006 Sharp, 2008). Accessing a high-paying STEM position
with a high salary was defined as career success in a career choice.
Predictive validity could increase accuracy if it predicted and measured the
criteria intended. The results correlated with other results (Gibilisco, 2011;
Kitchens, 2003, Moses & Knudsen, 2007; Nachmias & Nachmias, 2008;
Yin, 2003). For predictive validity, the goal of this study was to explore the
relationship between fixed gender roles discouraging participation in math
and science in high school and postsecondary school (Noddings, 1986). If a
female is not encouraged to take math and science in high school, the
chances of her taking STEM classes in postsecondary school are lowered
(Noddings, 1986). These goals were predicted and criteria measured through
the relations between the independent and dependent variables which are
participation in math and science and career choice which correlates with
career success (increased salary), respectively.
Empirical validity related to the Likert scales and survey instrument
and the results yielded as a result of the research. The convergent-
discriminate concept of validity stated that different measurements of the
same property should yield the same results (London, Rosenthal, &
Gonzalez, 2011; Yin, 2003). Using a Likert Scale should yield the same
results. This concept also increased consistency or reliability. The strengths
of these scales in relationship to the independent variable were that they
could be used to measure accurately, the number of years females
participated in math and science, and the number of STEM classes they
took, which were ordinal. These scales can also be used to measure their
attitudes towards these courses and how they participated, which are interval
(Nachmias & Nachmias, 2008). On the issue of norm or criteria referenced,
the Likert Scale was criteria referenced. Indirect criteria were those that
influence outcomes while not being linked directly or normally to the
validity of the variable. The pilot study helped determine the reliability of
the instrument by determining consistency in the responses over time.
Validity and Reliability
Validity is whether something measures what it purports to measure.
In other words, it was important to determine if a researcher could extract
meaningful data and inferences from the scores on the instruments
(Kaczmarek, Haladzinski, Kaczmarek, Baczkowski, Ziarko, & Dombrowski,
2012, McCullough, 2011; Shao, 2002; Sherman, 2004, Yin, 2003).Validity
showed if the survey was an effective one used to make inferences about a
population based on scales used and to determine if they measured what they
were intended to measure (Shao, 2002), making it very important. There are
also threats to validity which I must avoid in this study.
According to Shadish, Cook & Campbell,(2002) some threats to
internal validity in a study were if someone matures out of the study and
leaves, a subject in the study dies, or if the researcher was unable to establish
cause and effect between or among variables. Since this study was not a
longitudinal study, this threat did not apply to this study (Shadish, Cook &
Campbell, 2002).. This study was a cross sectional study. In a survey
method such as this study, the main threat to validity is the bias in the
wording of the questions and the bias in responses (Shao, 2002). To combat
this threat, I wrote the questions in an objective, unbiased manner using a
reliable 5-point Likert Scale. External validity was threatened when
incorrect information is inferred (Campbell & Stanley, 1963; Field, 2012;
Sherman, 2004). To combat this threat, I ensured an unbiased set of
questions without offering any additional information that can be incorrect.
In survey questionnaire construction, simplicity is the least bias (Nachmias
& Nachmias, 2008; Shao, 2002). To combat bias, I viewed the
histogrammes for obvious and cleaning out subtle outliers; use SPSS to find
the case causing the bias and verify the raw data (Field, 2013). Since my
sample is greater than 30 and is fairly large, I examined a normal
distribution as opposed to a skewed one.
Reliability was also a concern in this method. The scale was a 5-point
Likert scale using the following measurements; the proxy number of classes
females took, analysis background, both obtained through both the survey
and secondary data (Glass, 1976), and the last measurement is their current
position and salary. I used income categories to increase willingness of
respondents to answer salary questions (Bobbie, 2006; Shao, 2002). One of
my hypotheses is that there is a positive relationship between a career choice
in a STEM career and salary.
In the survey method reliability was also a major concern (Strauss &
Corbin, 1996). Reliability is consistency over time. In other words if the
researcher did a pilot study, then a pre-test and a post-test (Campbell &
Stanley, 1963) or a follow-up survey after a mass mailing or internet survey,
the latter in this case (Teijlingen,& Hundley
(2001), using the same instrument, the results would be consistently the
same (Shao, 2002; Sherman, 2004; Yin, 2003). If the instrument has a
Cronbach alpha of .7 or .8 after conducting the pilot, yielding similar results,
then the instrument is valid and reliable (Field, 2013).
Furthermore, according to Yin (2003) it is important to use a well
established scale rather than create one that has not been proven. For this
reason, I am using the 5point Likert Scale, that has been used many times
before (Shao, 2002; Yin, 2003), but with modified questions so that I could
specifically ask the research question. This was important because if a test
showed consistent results, that meant the instrument is credible and
consistent making the results useful (Shao, 2003; Yin, 2003).
Data Collection
The data were being collected by submitting to the alumni
associations how to access an online survey for the sample of alumnae for
each of the four sampled schools. The method was an online survey using a
5-point Likert Scale (Creswell, 2014; Shao, 2002). There were mostly
closed-ended questions which also included demographic questions to better
understand the difference between and among sampled groups (Kalton,
1983). The questions reflected the research questions of the study. Since this
was a simple online survey given by their alumni association gatekeepers,
there is no exit or debriefing process. This was a quantitative study,
employing an online survey, therefore, there is no interview, making a
debriefing process unnecessary. There was no follow up except for after the
pilot study (Yin, 2003).
Data Analysis
Linear regression determined the relationship between the interval
variables by expressing the relationship as an algebraic equation by
predicting outcomes, according to Nachmias and Nachmias (2008). The
residual sum of the squares indicated how well the line fits the data,
according to Field (2013).
When using non-experimental data, the variables were called the
predictor and the criterion, which was the same as the independent and
dependent, which I used the latter terms although my data were non-
experimental (Green, & Salkind, 2011). In the equation Y =: o +: 1 X1 +: 2
X2, Y was called the dependent or outcome variable, and X1 and X2 were
called the independent variables, or predictor variables
A multiple regression was more appropriate than other methods to
determine the relationship between the number of STEM classes taken and
role models and salary (Achen, 1982, 2009; Miles & Huberman, 1994;
Nachmias & Nachmias, ANOVA and a regression were possible. I could
determine the relationship between the variables using a two-sample T test
since the dependent variable is measured using an interval scale as well as
an ordinal one if the sample is small and if comparing means (Nachmias &
Nachmias, 2008). Furthermore, a multiple regression was conducted to
determine the relationship between the number of STEM classes taken and
number of same sex role models and salary, since this was a relationship
study and not a comparison between two means (Gibiliso, 2011; Nachmias
& Nachmias, 2008).
The more STEM courses a female took in high school and college, the
more likely she would choose STEM careers which tended to have higher
the salaries than most other fields. Career choice was a categorical (Field,
2013). For the quantifying of the career choices, there will be five groups of
career choices, reduced from the original seven where the measure is the
number of STEM courses taken, will be conducted with an ANOVA.
Subsequently, I determined using an ANOVA which career choices have the
highest average number of STEM courses. I could analyze the data by using
ANOVA to a group of career choices where the measure is number of
STEM courses taken (Nachmias & Nachmias, 2008). Then with an
ANOVA, I could determine which career choices have the highest average
number of STEM courses. I can analyze average salaries by career choices
and the average impact that role models have overall on salaries and career
choices. Then a post hoc will indicate which differences are significant with
a p value of less than .05 (Field, 2013; Nachmias & Nachmias, 2008).
For the salaries, a regression was performed to determine if the
number of STEM classes a female student takes was a valid predictor of
salary. Subsequently, I employed salary groups and these groups will be the
factor. The number of STEM classes is the measure and then I conducted an
ANOVA with the independent variable career choices, and the other factor
salary brackets, with the measure number of STEM courses, since the survey
is cross-sectional. The criterion of using alumni from four Long Island
universities who are born after 1980 are a nested classification in which an
ANOVA is effective, according to Campbell and Stanley (1963).
The data was screened and cleaned for outliers or extreme values that
can skew a distribution. Outliers could greatly impact the regression
equation (Field, 2013; Morrow, 2013). They could affect the precision of the
estimation of the regression weights. For this reason I dealt with this issue
with the independent variables and dependent variable, prior to conducting
my regression. A description of analyses used to detect differential attrition
or to ensure that groups are equivalent before the study is conducted. I will
search for outliers using a histogramme to view extreme values.
Descriptive Analysis of Data
A regression could answer research questions like how strong was the
relationship between (Field, 2013; Morrow, 2013) the independent and
dependent variable which for example, in this study was the number of
STEM classes a female takes and role models she has and the salary she
earns by choosing a STEM career. The regression can indicate the strength
of the relationship between these variables. The two predictor variables are
the number of STEM classes and role models and the outcome variable is
salaries for the regression in Research Question 3. This analysis can also
indicate which predictor variable has the strongest relationship with the
dependent variable.
There were several underlying assumptions for a multiple regression.
A researcher must be careful of outliers (Morrow, 2013). These outliers
could affect the precision of the estimation of the regression weights,
making data cleaning necessary before the regression is conducted. The data
cleaning was conducted by deleting any outliers (Morrow, 2011). The next
assumption was ratio of cases to predictors or independent variables.
Regression could be sensitive to sample size. If the sample was too small,
the researcher will not obtain an accurate prediction equation of the
independent variables to the outcome variable (Kalton, 1983). To be able to
accurately test for the multiple correlation, and each of the individual
regression coefficients, the sample size must be at least N greater than or
equal to 104 plus M, where M is the number of predictors or independent
variables in the regression (Field, 2013; Morrow,
2013).
Next, like an ANCOVA, the regression was sensitive to
multicollinearity, which is when there are least two predictors or
independent variables, in the equation, that are too highly correlated with
each other (Field, 2013; Morrow 2013). Multicollinearity could reduce the
reliability of the regression and could create large standard errors in the
equation. The next assumption is the normal distribution of variables
without skewness or kurtosis (Field, 2013). The prediction equation is
enhanced if the variables are normally distributed. For any linear model to
be valid it must be assumed to have additivity and linearity (Field, 2013). It
is also assumed that for any two observations that the residuals should be
independent and not correlated. This assumption can be tested with the
DurbinWatson test which tests for serial correlations between and among
errors. Generally values between 1 and 3 are problematic (Field, 2013).
All multiple regressions have homoscelasticity. This meant that
residuals at each level should have the same variance. This goes for all linear
models which include multiple regression. Furthermore, for all linear
models, it is assumed that the residuals are random, normally distributed
variables with a mean of 0. Moreover, predictor variables are uncorrelated
with external variables which are variables that have not been included in
the regression model that influence the outcome variable (Field,
2013).Lastly, the predictor variables should have some variation in value,
meaning they cannot have a variance of 0.
I used the standard multiple regression for Research Question 3,
which is the most commonly used which is the one I will be using (Morrow,
2013; Nachmias & Nachmias, 2008). In this type, all of the predictor and
outcome variables are entered into the linear equation simultaneously. Each
predictor is assigned only its unique variance that it contributes to the
equation. Variance referred to the amount of overlap the predictor has with
that outcome. None of the predictor variables were assigned the overlapping
variance which is the overlap that is shared among these predictor variables.
The overlapping variance still is part of the adjusted R-squared, but it was
not assigned to an individual predictor variable (Kitchens, 1983; Morrow,
2013). This type needed at least 104 plus M participants but it needed the
lowest amount of all the types of multiple regression (Morrow, 2013). The
change in the R2 statistic is produced by adding or deleting an independent
variable. If the R-squared change associated with a variable is large, that
means that the variable is a good predictor of the dependent variable.
Hypothesis
The hypothesis was that the more STEM classes females take and the
same sex more role models a female has, the more likely she is to choose a
STEM career and also receive a higher salary. In the first two questions, the
idea is that in retrospect, females who choose STEM careers tended to take
more STEM classes and have more same-sex role models. The gap was that
no study has focused on females born after 1980, living in Long Island
suburbs that still drop out or avoid STEM classes and careers because of the
desire for a flexible career, with less challenging academics, due to a lack of
role models (Farland-Smith, 2009). The STEM courses taken in school, and
the career choice of a STEM career, could result in higher pay which
correlates with career success (Field, 2013; Nachmias & Nachmias, 2008).
For Model Building Strategies for Regression
There were three basic different types of regressions that I can use in this
study. The first one is a standard multiple regression, which is the most
commonly used and the one that I used (Duntemen & Ho, 2006; Morrow,
2013; Nachmias & Nachmias, 2008). In this type, all of the predictor and
outcome variables were entered into the linear equation simultaneously.
Each predictor is assigned only its unique variance that it contributes to the
equation, which is the amount of overlap the predictor has with that
outcome. None of the predictor variables are assigned the overlapping
variance which is the overlap that is shared among these predictor variables.
The overlapping variance still is part of the adjusted R-squared, but it is not
assigned to an individual predictor variable (Duntemen & Ho,
2006;Kitchens, 1983; Morrow, 2013). If the R-squared change associated
with a variable is large, that means that the variable is a good predictor of
the dependent variable. My sample was originally more than 104 but since I
did not I receive cooperation I needed from the universities, thus, my sample
size was substantially reduced to a small data set of 48, due to low response.
I only had two predictor variables which are the number of STEM classes
females took and the number of female role models in questions 3, but these
variables are dependent variables in questions 1 and 2. Thus, it was not
necessary to use a stepwise approach the standard method may suffice,
unless I wanted to see which variable has the closer relationship with career
choices (Field, 2013; Morrow, 2011).
The stepwise or sequential or hierarchical is where the order is
dependent on prior theory or research. As the researcher, I would enter the
predictors in an order that I would specify. A researcher can enter each
predictor individually, at each step, or he or she can enter sets of predictors
at each step in the regression equation. Overlapping variance was assigned
to the predictor variables (STEM classes and role models) in the order that I
would enter them into the regression equation (Field, 2013; Hamburg, 1983;
Morrow, 2013). In this case, the order that one entered the independent
variables into the equation was contingent on statistical criteria (Field, 2013;
Morrow, 2013), where SPSS can decide which order the independent
variables are entered based on the statistical criteria that the researcher
would enter. Each predictor is given its own unique and overlapping
variance when it is entered into the regression equation (Nachmias &
Nachmias, 2008).
In the stepwise or stepping method these options apply when either
the forward, backward, or stepwise variable selection method has been
specified. Variables could be entered or removed from the model depending
on either the significance (probability) of the F value or the F value itself
(Field, 2013). The stepwise method is the same as the forward method.
According to Field (2013), there is the constant b(0) where the computer
decides the predictor variable order based on what is left by looking for the
variable that can explain the largest percentage of the outcome. The
backwards method is the opposite of the forward. The computer places all
the models’ predictors and observes the significance values. There is a
removal criterion that if a predictor variable does not have a significant
impact on the outcome, it is removed. If I wanted to see whether STEM
classes that females took or female role models have a closer relationship to
career choices, then I could conduct the regression in a stepwise fashion
both forward and backward. However, this was not necessary for this study.
In the hierarchical or blockwise entre was based on past work and the
researcher decides what order to enter the predictor variables into the model,
according to Field (2013).The variables should be entered in their
importance in predicting the outcome. In this method, the researcher can add
new predictors. This one did not apply to my study.
The forward method was used when the independent variable that has
the largest bivariate correlation with the dependent variable is entered into
the regression equation first. I did not need to try this method because was
already evident ‘that STEM classes’ was a better predictor of higher salaries
than ‘role models’ using the linear regression model. I will examine the
relationship between the independent variable career choices and the
dependent variable STEM classes using an ANOVA in Research Question 1.
The second method is called the backward method. This was when the
independent variable has the smallest bivariate correlation with the
dependent variable and is entered into the equation last (Field, 2013;
Morrow, 2013). When a researcher used a simple multiple regression or a
statistical one, this has an impact on the total N solution, causing it to differ
if the researcher uses one method or the other. Using a CI of 95%, there is a
95% chance that the mean is between the lower and upper bound meaning a
95% chance that the population mean is included.
Descriptive and Inferential Statistics Reported
The statistics that were reported once the data collection from the
surveys were conducted and analyzed was the output from the ANOVA for
the number of STEM classes as the dependent variable in Research Question
1, and career choices as the independent variable. This ANOVA included
additional post hoc hypothesis tests that must be managed once the ANOVA
has been conducted. The post-hoc tests were implemented after the repeated
measures ANOVA or factorial ANOVA to determine mean difference,
significance or non-significance in the p value, which is significant at less
than ,05 (Gibilisco, 2011; Green & Salkind, 2011; Morrow, 2013). The
Schefflé test is a conservative test that compares all pairs of means. The
more progressive test is the Tukey HSD test which also compares all the
pairs of the means (Gibilisco, 2011; Green & Salkind, 2011; Morrow, 2013).
The test I used was the LSD post hoc test.
Subsequently, the effect size was generated by figuring out the percentage of
variance, which uses the formula of the sum of squares between divided by
the sum of squares total (Morrow, 2013). Also one effect is interacted with
another (Field, 2013). There is also the Levene test which is used if the
violation of homogeneity is violated, but it was not violated in this study, so
this test was not necessary. However, since this test only matters with
unequal group sizes, this test is only used in that case. This test was
irrelevant with equal group sizes (Field, 2013).
Statistics included descriptive statistics such as the mean, and standard
deviation for each of the variables, with an LSD post hoc correction test.
The multivariate tests will include the F ratio, the degree of freedom for the
hypothesis and the error, the mean squares, the significance of the
relationship of each predictor variable to the outcome variable, and the
partial eta squared (Field, 2013; Morrow, 2011; Nachmias & Nachmias,
2008).
For the regression, the objective was to see how close the relationship
STEM classes and role models are to career choices or salaries. This will
answer the research question. The hypothesis test was an extension of the t
test. The statistics that are demonstrated in this analysis are similar to the
ANOVA (Field, 2013). The ANOVA was used to quantify the categorical
variable of career choices. Here the statistics included descriptive statistics,
including the mean and the standard deviation for each variable for this
original sample of 487, reduced to a small data set of 48. There will also be a
Pearson’s correlation, and the significance in the relationship between each
predictor and the outcome variable (Field, 2013; Kitchens, 2003). Then the
model summary showed the R statistic, the R squared, adjusted R squared,
the standard error of the estimate, the degrees of freedom, and changes in R
squared and the F ratio. The adjust R squared was the overall effect size for
the multiple regression. The R-squared tended to be an overestimate. There
was also the ANOVA summary table with the same statistics as indicated by
the ANOVA (Morrow, 2011). For the coefficients, there was the
unstandardised B which encompassed both the weights and the standardized
which were Beta. The unstandardised coefficients are B weights, which
represented the slope, keeping all else constant. The beta weights were the
standardized coefficients. The larger the beta weight, the stronger the
relationship between the independent variables and dependent variable
(Field, 2013; Morrow, 2011). This also included the confidence interval of
95%. Using a CI of 95%, there is a 95% chance that the mean is between the
lower and upper bound meaning a 95% chance that the population mean is
included. Then there are the Collinearity diagnostics which include variance
proportions for each variable.
Multicollinearity was when there are least two predictors or
independent variables, in the equation, that are too highly correlated with
each other. In this case the number of STEM classes and the number of role
models were correlated but not highly correlated. The correlation was
positive but weak for role models and more significant for STEM classes.
Also, they were not be highly correlated with participants’ salaries.
However, this assumption is the same for all linear models. There should be
no perfect linear relationship. This multicollinearity could reduce the
reliability of the regression and can create large standard errors in the
equation making it difficult to assess how close the relationship was between
both STEM classes and role models to career choices.
Power included sample size, B, effect size, and C, alpha level. A post
hoc analysis was computed as a function of the population effect size
parameter, and the sample size used in a study (Faul, et al., 2009). The alpha
level is the chance of error that researchers are willing to take in determining
statistical significance. An alpha level at .05, means the willingness to accept
a five percent chance of error in their statistical analysis, if H0 is correct and
rejected, which goes along with a confidence interval of 95% (Field, 2013;
Morrow, 2011; Sheperis, 2014). If one uses .10, the chance of finding
significance increases. Power should be set at .80 which means that as a
researcher I have an 80% chance of finding a significant difference between
my variables, and only 20% chance of committing a type II error. The effect
size is the impact of any treatment or intervention or the impact of the
number of STEM classes and role models on career choices. To calculate the
G power analysis, the effect size is medium at .50, which is acceptable for
Cohen’s D as postulated by Field (2013) and Sheperis (2014).
Furthermore, the alpha or significance level is .05 and these
calculations were for a two-tailed test. The error problem is .05, and 1 – b
is .80. N = 487, for the original sample before it was reduced. Power is one –
beta and this helps to avoid type II errors where a researcher failed to reject
the null hypothesis (Sheperis, 2014). To prevent these errors, this calculation
employed the output non-centrality parameter represents the degree to which
the null hypothesis is false, so that type II errors can be prevented. The
output parameters also provided critical value, degrees of freedom, and the
sample size for each group of sampled participants, the total sample size,
and the actual power for the study.
Here the sample size for my study was estimated. This is a priori
power analysis which computes the required sample size, given alpha level,
power, and effect size as indicated. My effect size turned out to be .15 which
is low to medium. My power 1 – b was set at .80 and my alpha is .05. I
conducted a linear regression fixed model, single regression coefficient. I
also conducted a priori which computed sample size, effect size, power, and
alpha. My non-centrality is 3.693 my critical T is 1.9789, degree of freedom
is 86 and my sample size was 122 before the reduction to 48 which was the
number of alumnae I surveyed randomly at the four universities in my
sample. My actual power is .95 and my partial r squared is ,5 with a residual
variance of 1.
The effect size is how strong the relationship is between the numbers
of STEM classes taken and career choices in the field. The effect size is the
mean difference over the standard deviation (Burkholder, 2010). In this case
if alpha or significance level is .05, and power is .80, then determine the
effect size. The power was the odds that one can observe a treatment effect
when it occurs or the odds of saying that there is a relationship, difference,
gain, when in fact there is one (Trochim, 2006d). A lower alpha makes a
type one error less likely. I used Cohen’s d. It was a good idea to include a
buffer for attrition for refusals to participate in the survey (Shao, 2002).
Data Analysis Plan
Rationale for Methods Not Used
Chi Squares. The Chi Square is a nonparametric test that evaluates if
the actual proportions of individuals who fall into a category are the same as
the hypothesized version (Field, 2013; Green & Salkind, 2011). They are
used when testing hypotheses of equal and unequal proportions (Green &
Salkind, 2011; Nachmias & Nachmias, 2008). Chi Squares are used for
cross-tab analysis. Chi Squares offer goodness of fit and tests of
independence (Hamburg, 1983). Chi Squares provided the basis in which to
judge whether or not two population proportions are equal (Hamburg, 1983).
Cramer’s V assesses the strength between row and column variables and
ranges from 0 to 1. A phi coefficient is used for a 2x2 tables and ranges from
+1 to -1 from strong positive to strong negative (Green & Salkind, 2011;
Morrow, 2011). Values close to 0 signify a weak relationship or no
relationship and non-significant (Nachmias & Nachmias, 2008).
If the difference between the frequency observations and frequency
experience under a set of assumptions are significant, two nominal variables
in a cross tabulation can be used. However, these variables of math and
science courses females took are ordinal variables using ranking. In addition,
in this study I employed ordinal and interval scales to measure ordinal
variables. A chi square could be used because the variables must be nominal
(Green & Salkind, 2011; Hamburg, 1983; Kitchens, 2003; Morrow, 2011;
Nachmias & Nachmias, 2008).
Bivariate Analysis
Bivariate analysis is the analysis between two variables using cross-
tabulation (Nachmias & Nachmias, 2008). Since in this study I used three
variables in Question 3, this analysis cannot be used (Field, 2013; Nachmias
& Nachmias, 2008). In this case either a t test or a multiple regression would
be necessary. Bivariate analysis is using paired tables, related to regression,
and variation between the means, which were not being compared in this
study (Nachmias & Nachmias, 2008). There was a cross tabulation using a
two variable table to analyze the relationship between the dependent and
independent variables (Green & Salkind, 2011; Nachmias & Nachmias,
2008). The other statistics included the residuals which are errors.
Test
The assumptions of the t test were that the sample observations must
be independent (Shadish, Cook, & Campbell, 2002). ANOVA is the
extension of the t test. The distribution is close to normal the closer the
sample population is to 30 or more than 30. My sample is 48. The
population must be normally distributed, which has one hump and is not
skewed (Field, 2013; Gibilisco, 2010; Morrow, 2010). If N ≥ 30, I can use
either the z or the t test but I did not compare two samples. However, if the
distribution is normal, this would be a possibility. One sample T tests were
used for single samples, or paired or two independent samples. They are
distinguished by the choice of the test variable. These include the midpoint
of the test variable, its average value based on past research, and its changes
in performance (Green & Sulkind, 2011). The t test evaluates whether the
mean of the difference between the independent and dependent variables are
significantly different from zero using a repeated measure or matched
subject design (Kitchens, 2003; Green & Salkind, 2011; Nachmias &
Nachmias, 2008). A t test works well with a large sample since the t scores
are almost the same as the z scores, when the sample is this large. In this
case, the distribution was close to normal, and this reduced standard error
(Kitchens, 2003; Nachmias & Nachmias, 2008).
For this study, I did not use the two sample t test because I did not
compare sample means, but the one sample is possible. Instead, I compared
five groups of career choices employing an ANOVA. This test applies to the
comparison data extrapolated about both of these random samples. The T
test works well with such a large sample since the t scores are easily
transferred to z scores and the distribution is close to normal (Field, 2013;
Kitchens, 2003; Nachmias & Nachmias, 2008).
A two sample paired t test could be conducted in order to determine if
the mean difference between the variables, significantly different from zero.
Since the sample size is large, that means N ≥ 30, I would need to convert to
Z scores. For a two tailed test, the Z score which is the normal distribution
goes from ± 1.65 and ± 1.96 for a significance level of .05 and ± 2.58 and ±
2.33 for a significance level of .01 (Nachmias & Nachmias, 2008). The t test
evaluates whether the mean of the difference between the independent and
dependent variables were significantly different from zero using a repeated
measure or matched subject design (Gibilisco, 2011; Green & Salkind, 2011;
Kitchens, 2003).
This is conducted after hypothesis testing. If one were to use this method,
one must then go to table and look at the type of test, two or one tailed test,
and the critical value to see if t test is significant (Gibilisco, 2011; Green &
Salkind, 2011; Kitchens, 2003; Nachmias & Nachmias, 2008). The effect
size is the mean difference over the standard deviation were devaluates the
degree that the mean scores on the test variable differ from the value in the
units of the standard deviation (Green & Salkind, 2011).
If p is assumed to be zero under the null hypothesis, I could test the
statistical significance of r to a standard score using the t statistic with an n-2
degree of freedom. The formula is: √ is square root. Formula t = r√n-2/√t-r2
(r squared). The null hypothesis stated that there was no relationship
between math and science courses taken by females in this sample and their
career choices and the salaries for female in the sample. The hypothesis says
there was a direct or positive relationship between the STEM courses taken
by females and choosing STEM careers. Committing a type one error would
be if the hypothesis is true and I rejected it. A type II error would be if I
accepted the hypothesis and it is false. These are errors that I must avoid in
my conclusion of whether or not I accept or reject the hypothesis (Nachmias
& Nachmias, 2008). Therefore, I must look at the significance of the
relationship between the number STEM courses taken at the high school and
postsecondary level and the career choices made inn STEM fields. The
significance is between .01 and .05 which means that the null hypothesis
stating that there is no relationship between these variables would be
rejected, the resulting sample would have occurred randomly no more than
one percent or five percent of the time, according to Nachmias and
Nachmias (2008). The level of significance at 5% if the null hypothesis was
rejected would be that 5% of the true hypothesis has been tested.
Coding of Survey Responses
From the survey instrument and the three interval and ordinal scales,
each response were coded with a number (Miles & Huberman, 1994;
Nachmias & Nachmias, 2008; Patton, 2009). Code numbers offered rank
and order (Shao, 2003). The Data were coded and compared with records
that have undergone an aggregate analysis to find any relationships (Babbie,
2006). The names of math and science courses were exhaustive. If the
respondents have taken these classes, it would be exclusive and what
classes, levels, proficiencies, and interests are detailed (Nachmias &
Nachmias, 2008; Orr & Mitchell, 1994). The coding was a computation of
the sum of the codes for each response. In a 5point Likert Scale, response
one is a one, two = 2, three = 3, four = 4, and five = 5. These values were
added and I determined the discriminate power by determining the highest
and lowest value responses. Then I selected the highest discriminate powers
and test the reliability (Nachmias & Nachmias, 2008).
To avoid errors, the data was directly input into my laptop in SPSS,
without the use of transfer sheets or edge coding (Miles & Huberman, 1994;
Nachmias & Nachmias, 2008). In addition, the salary groups will be
correlated with the respondents’ career choices and math and science classes
taken in high school and postsecondary school which correlate with salary.
Then, I ran a regression to analyze these relationships using SPSS.
Limitations and Delimitations
Limitations of the Study-Design Weaknesses
The potential design and/or methodological weaknesses of the study
were for one, that this study is using STEM career choice, salaries, and if
they relate to math and science classes taken in high school and
postsecondary school, as well as number of female role models, and
information on classes females took may be difficult to obtain. Therefore,
this secondary data were difficult to obtain due to confidentiality, making it
difficult to mitigate this issue (Glass, 1976).
Also, a major limitation of the study was the sampling strategy. It
was difficult to obtain the sampling frame of alumnae due to confidentiality
issues (Kalton, 1983). Two phase sampling may be necessary to obtain the
list for this sample (Kalton, 1983). Contacting alumni associations can be
difficult since they may be unwilling to contact their students for a study
due to confidential concerns. To mitigate this situation, I assured the
alumni associations, who were the gatekeepers (Campbell & Stanley,
1963), that the IRB of the university insures confidentiality, dignity, and
anonymity for all participants. The fact that the study is online is both a
weakness and strength. It is a strength because it is fast, and cheap or low
cost to administer, mitigating the cost weakness. However, it is a weakness
because not everyone has access to the Internet (Case, 2007). Since most of
the respondents are from the four Long Island schools or the Walden pool,
it was most probable that the female alumni in the sample frame will have
internet access (Case, 2007).
Another limitation of the sampling frame was missing elements where
it makes it possible that a particular female alumna comparison has no
chance of participating in the study (Kalton, 1983). With an internet survey,
it was difficult to make it completely random because of the issue of lack of
access, but again, since the sample frame was extracted from the alumni
association gatekeepers for the students, this increases chances of internet
access (Case, 2007).
Threats to Validity-Weaknesses
According to Shadish, Cook & Campbell, (2002), prior knowledge of the
instrument and testing can also bias a study and is a threat to internal validity
(Becker, 1986; Field, 2013). Therefore, one disadvantage to the survey was
that they do not control extraneous variables efficiently and they have a
difficult time controlling threats to internal validity and generalisability.
According to Ahern (2005), the internet was less expensive; it saves time,
makes information globally accessible, can improve external validity or
generalization, and has increased access to information on sensitive issues
and for special populations, which counteracts some of the difficulties
surveys have in controlling threats to validity. This is one of the issues I
faced with the pilot, according to Campbell & Stanley (1963).
The respondents could have prior knowledge of the test. To mitigate
this problem, I will let time lapse to give respondents a chance to forget.
Specific events occurring between these two time-lapsed tests can also bring
in other variables that could impact responses (Campbell & Stanley, 1963).
There is also the effect of taking a test upon the scores of a second testing,
according to Campbell & Stanley (1963). The other issues of internal
validity that I dealt with were from biases from the selection of my sample,
maturation or morbidity from the study or statistical regression. This was
when I coded the Likert Scale and selected the highest discriminate powers
and select groups or individuals based on their extreme scores. This is a
threat to internal validity.
To ensure content validity, I made sure that the information and data
from the literature were properly cited and accurate (Field, 2013; Gibiliso,
2011; Nachmias & Nachmias, 2008). The convergent-discriminate concept
of validity stated that different measurements of the same property should
yield the same results. A weakness would be if it does not yield the same
results. If that occurs, another test may be required. A weakness of empirical
validity would be if the research is not accurate. Through triangulation,
using the Likert Scale at the levels of both interval and ordinal, plus with
employing a survey and an ANOVA, I did my best to ensure accuracy.
Other threats to internal validity included regression, the personal
experience of the participant, the procedure of the relationship, or the subject
being measured (Nachmias & Nachmias, 2008; Shadish, Cook & Campbell,
2002). While I could not control for personal experience of the participant,
recognizing this fact and mentioning it in the limitations of the study, added
validity to the findings (Yin, 2003). I could only control for the bias in my
own personal experience, which I did using these scales and making sure the
survey questions are objective (Trochim, 2006a). This was inherently a
biased topic, for this and other reasons; a quantitative approach was
employed with three reliable and valid scales to negate this issue (Case,
2007; Shao, 2002; Yin, 2003).
To attempt to reduce the threats, the survey was simple, clear,
confidential, online, and convenient to increase response rates (Patton, 2009;
Watkins & Corry, 2007; Yin, 2003). There were also statistical analyses
used such as the regression (Green & Salkind, 2011). The questions were
engaging, objective, and the demographic questions will be at the end (Shao,
2002). Threats to validity such as prior experience of the participant could
not be reduced but by recognizing them, but this could be recognized as to
the study. However, my personal biases were reduced by trying to remain
objective (Kitchens, 2003; Miles & Huberman, 1994; Nachmias &
Nachmias, 2008; Onwuegbuzie & Leech, 2005). The subject matter is
inherently bias, but by keeping the questions as objective as possible, it
reduced threats to validity (Shao, 2002). Threats to external validity were
inaccuracies in data or incorrect data. Therefore, the data were not
generalisable. In order to reduce external validity, it was important to ensure
that the research design is sound, the data are entered accurately and it is
analyzed accurately (Kitchens, 2003; Nachmias & Nachmias, 2008). Data
was coded and entered slowly and accurately and checked before it was
entered (Miles & Huberman, 1994).
Threats to External Validity
The question of external validity was generalisability. Some of these
factors, according to Campbell & Stanley (1963), were the issues of
representativeness and the ability to generate my findings to the entire
population. The issue was if the findings could be generalized for all males
and females and not just the sample in Long Island from these four
universities. There was a reactive or interaction effect of testing was a
pretest or a pilot study, according to Campbell & Stanley (1963) and Yin
(2003). This meant that the pretest or pilot may have decreased the
respondents’ responsiveness to the dependent variable which is the number
of STEM classes in Research Question 1 and the number of same-sex STEM
role models in Research Question 2. There could be an interaction between
selection bias and this variable. In order to mitigate these issues, the
gatekeepers from the alumni associations will randomly choose the alumni
participants to offer some randomness to reduce bias. Another way to
mitigate the pilot issue is by using different respondents for the pilot than I
use for the regular study. This would also mitigate the threats of internal
validity.
Ethical Procedures
The data will be destroyed five years after publication of the study and
all responses are confidential, private, and anonymous. The Institutional
Review Board (IRB) ensures the ethical treatment of human subjects. The
Institutional Review Board review reference number is 11-18-15-0169928.
The researcher will have no direct contact with the subjects since the
instrument will be distributed online through the alumni association
Summary of Chapter
Here was the introduction to the methodology of data collection and
analysis and the justification for the research design chosen for this study.
Moreover, the research questions were related to the survey questions. The
use of a simple stratified random sample was used and the population was
discussed. Issues of confidentiality as well as informed consent are very
crucial. The instrument was measured for validity and reliability.
Descriptive and inferential statistics were analyzed in this chapter as well
Chapter 4: Results
Introduction
The interest in STEM fields is often diminished at some point prior to
postsecondary educational level for some females. It seems to wane around
middle school or early high school due to lack of exposure or preparation
(You, 2013). As mentioned in earlier chapters, this also may be due to
perceptions that these fields are too difficult and a lack of family
encouragement when females are little from parents, teachers, and family
(Buschor, Berweger, Keck, & Kappler, 2014). Therefore, females may
believe that these fields are geared towards the so-called mechanical
aptitudes of males and other such stereotypes, which may have resulted in
females avoiding STEM majors and careers. For this reason, it is imperative
to encourage girls when they are young to be interested in STEM and to
invest in programs that would encourage them (Wang, Degol, & Ye, 2015).
Perhaps such encouragement will increase retention in STEM even past
middle school, into high school, postsecondary, and into their careers
(Drane, Micari, & Light, 2014; Gershenfeld, 2014). Therefore, as a
conclusion, peer and faculty mentoring programs for young females
employing female role models should be initiated in all school districts, at
all grade levels. The purpose of such mentoring is to encourage females to
take more STEM classes, retain female interest in STEM, and become more
interested in STEM at postsecondary level to encourage choosing STEM
careers.
One recent example of such a mentoring program is Project Scientist,
based in North Carolina. This program was one that is mainly targeted at the
elementary and middle school age females (Polk, 2014). The Project
Scientist program is significant because it becomes the bridge for students to
create an interest in STEM and to change society’s perception in female’s
interest in STEM. According to Polk (2014), 78% of school-aged young
females have an interest in STEM, yet adult females only make up 25% of
the STEM workforce, as mentioned in earlier chapters (Department of
Commerce, 2009). Project Scientist provided support with female STEM
role models to develop an educational plan for females to be better equipped
to choose STEM careers. For this reason, I have collected data from the
alumnae of several LI universities to determine their experience with STEM
classes and careers through an online survey. Moreover, I obtained
additional responses from the participation pool at Walden University.
Data Collection and IRB Results
The data were collected from a sample of four universities on Long
Island and the participation pool at Walden University. For the four
universities, I was able to obtain cooperation from the alumni associations as
mentioned in the Institutional Review Board review reference number 11-
18-15-0169928 and the participation pool in the revised IRB change
approval. The reason for the addition of the participation pool was to
stimulate additional responses. The method of data collection was the online
survey where a link using a free service called typeform was provided to the
sampled universities’ alumni association for the alumnae to access. The
alumnae accessed the survey directly by clicking
https://edith11.typeform.com/to/EK2EVo and proceeding to answer the
questions. Furthermore, to increase responses, I expanded the research pool
to some additional universities and STEM business groups. I conducted a
small pilot study testing the survey before the research began because I
needed to assess whether or not I needed to modify the questions if
necessary to answer my specific research questions (Teijlingen & Hundley,
2001) in order to validate the survey. The original categories from Chapter 3
were divided into seven for the analysis of variance. Since I used small data
sets that can only handle five categories, I reduced the categories from seven
to five. This occurred due to the low response rate and reduced cooperation
from the universities. Originally, the categories or factor groups as indicated
in Chapter 3 are defined statistically below. µ1 = career choices for females
in science µ2 = career choices for females in technology/IT µ3 = career
choices for females in engineering µ4 = career choices for females in math µ5
= career choices for females in caring professions µ6 = career choices for
females in education µ7 = career choices for females in nontechnical fields
like legal, business, etc
However, the datasets are small and can only handle five factor
groups or categories which have been reduced to
A= 1 = µ1 = career choices for females in science and math
B= 2 = µ2 = career choices for females in technology/IT
C= 3 = µ3 = career choices for females in engineering
D= 4 = µ4 = career choices for females in nontechnical fields like legal,
education, soft sciences like political science, economics, business, etc
E= 5 = µ5 = career choices for females in caring professions or
nonprofits
Collection of Data
The method of data collection was a survey using a 5-point Likert
scale to measure the factors using single measurement and weighted average
for the ANOVA and the linearity of the regression. The sample size was
considerably smaller than I had hoped. Originally, I had forecasted an effect
size for a sample size in the 400s, and I only received responses from 48
including the four from the pilot respondents. For this reason, I have very
small data sets negatively impacting on the external validity.
Pilot Study
Prior to commencing the actual data collection, I conducted a small
pilot study of 4 out of the 48 respondents, testing the validity of the survey
using the Cronbach Alpha before the research began (Teijlingen & Hundley,
2001) because I am the original designer of the instrument. The pilot helped
me determine if I needed to modify any of the questions to increase validity.
The pilot did not appear to warrant any major modification in the
questionnaire, and I did not deem it necessary to perform a post test
(Nachmias & Nachmias, 2008; Teijlingen, & Hundley, 2001; Yin, 2003).
According to Campbell and Stanley, (1963), a pilot increases validity and
reliability. For the pilot, I used four of the first 48 responses just to test the
instrument, and while I viewed these first four, I noticed a few minor issues
with the instrument where on Question 20, I needed to add a none of the
above category to increase validity. Also on this same question, I had to
correct a minor typo. Moreover, one of the respondents in the pilot study
suggested that I add the time it takes to fill the survey in the introduction,
which I did. The average time estimated to complete the survey was reported
as 10 minutes, but the pilot showed an average of 6 minutes completion. By
doing these minor corrections and increasing my marketing and promotions
of the study, I was able to increase the response rate. The pilot results were
combined with the rest of the results since it was small and there were no
significant differences.
The pilot appeared to support the hypothesis that the relationship
between choosing STEM careers increased with the number of STEM
classes and role models. One respondent said she was not encouraged to
take math and science as a young girl and had no female role models. This
appeared to be consistent with the literature and the hypothesis as well as the
theory of Erikson (1980) stating what happens early in life affects what
occurs later in life. This was only a pilot consisting of four responses;
therefore, I could not draw conclusive evidence based on such a small data
set. For this reason, I combined the pilot with the other responses in all data
analyses. One woman in the pilot said that when she was little, she enjoyed
playing with dolls and caring roles. Therefore, she did not take many STEM
classes and she chose a nontechnical career, which also seemed to
synchronize with Erikson’s theory. Moreover, like the literature indicated,
interest in engineering among the pilot respondents was nonexistent.
Data Collection and Conversion of Data
The data were collected by submitting a link to the alumni
associations so that the alumnae could access an online survey directly for
each of the four sampled schools and the additional STEM groups and
Walden participation pool that were added to increase participation. For
Walden, I posted the link for Walden graduate students who had prior
degrees and, therefore, were alumni of various institutions. The method was
an online survey using a 5-point Likert scale. Subsequently, I asked closed-
ended questions that included demographic questions to better understand
the difference among sampled groups (Creswell, 2014; Kalton, 1983). The
questions reflected the research questions of the study. Since this was a
simple online survey given by their alumni association gatekeepers, there
was no exit or debriefing process. This was a quantitative study, employing
an online survey; therefore, there was no interview, making a debriefing
process unnecessary. There was no follow up after the pilot study (Yin,
2003).
The response rate was for 48 responses, of which the first four were
the pilot, and was out of 72 actual visits; therefore, the response rate for the
first group of responses was 42%. The responses came from laptops or PCs,
smart phones, and tablets. Approximately 38% of the responses came from
laptops and PCs. The average time of completion for the actual study’s first
group of respondents was the same as the pilot, which was 6 minutes.
The data for the all the responses were first input into Excel templates
before they were analyzed using the ANOVA and the multiple regression
(Field, 2013). However, then I used SPSS where I obtained more accurate
results with both the ANOVA and the multiple regression. Moreover,
according to Corder and Foreman (2014), it is important to ensure that the
data extracted from the analysis of variance and the regression (Aczel &
Sounderpandian, 2009; Field, 2013), which are both parametric tests, have a
goodness of fit. Since parametric tests were the optimal tests to use for this
application, nonparametric tests were not used.
The data from the participation pool were analyzed along with the
original sampled data from the schools even though the data from the
schools was from a stratified random sample whereas the data from the
participation pool was self selected imposing a bias (Field, 2013; Morrow,
2011; Nachmias & Nachmias, 2008). However, since the participation pool
responses were only around four or five, the bias was insignificant,
rationalizing why I was able to analyze all the data together. Moreover, since
the difference with the pilot was insignificant, I also analyzed the data with
the pilot and conducted one analysis of all 48 responses.
Survey Participants’ Demographic Classifications
The participant demographics were females who were alumnae of
either a bachelor’s degree or master’s degree program in a university
on Long Island or the Walden pool of participants. They are all
females born after 1980 from all income levels.
Missing Data
I took into account outliers and missing data to determine residuals
and standard error. Each set of residuals was independent of previous
observations. Any residuals that were not independent were considered self-
correlated. These outliers can affect the precision of the estimation of the
regression weights, making data cleaning necessary before the regression is
conducted. The data cleaning was conducted by deleting any outliers
(Morrow, 2011). Moreover, if a respondent did not fill out the number of
role models or neglected to answer the question, then it was assumed that
respondent did not have any role models and that the individual was self-
motivated, particularly if she answered interested or very interested in the
first few questions when asked about degree of interest in STEM.
Descriptive Analysis of Data
In analyzing the data, the two basic methods of data analysis
employed were an analysis of variance for the first two research questions
and employing a multiple regression for the third research question. For
Research Questions 1 and 2, the independent variable was career choices
and the dependent variables were the number of STEM classes and female
STEM role models, respectively. For this reason, two separate ANOVAs
had to be conducted, one for career choices and number of STEM classes to
answer Research Question 1, and the other for career choices and the
number of role models to answer Research Question 2. I hypothesized that
the more STEM courses a female took in high school and college, the more
likely she would choose STEM careers, which tend to have higher the
salaries than most other fields; therefore, her career choice would influence
the STEM classes she would take. Career choice is a categorical (Field,
2013). For the quantifying of the career choices, there were originally seven
groups of career choices where the measure is the number of STEM courses
taken that would be conducted with an ANOVA. However, I collapsed the
factors based on the responses since some groups had very small response
rates, as indicated in the previous section. The five groups or categories for
career choices were coded as follows: A or 1 for math and science, B or 2
for IT, C or 3 for engineering, of which there were no responses in this
category, D or 4 for nontechnical, and E or 5 for the caring professions.
Subsequently, I determined using an ANOVA, where career choice
was the independent variable and the dependent variable was the number of
STEM courses. I analyzed the data by using ANOVA to a group of career
choices, where the measure was the number of STEM courses taken
(Nachmias & Nachmias, 2008). Then, with an ANOVA, I determined which
career choices had the highest average number of STEM courses. I analyzed
average salaries by career choices and the average impact that role models
had overall on salaries and career choices. Then a post hoc indicated which
differences were significant with a p value of less than .05 for STEM classes
and salaries (Field, 2013; Nachmias & Nachmias, 2008). For role models
and salaries, the relationship was weaker than for STEM classes. The first
set of tables and figures and data analysis reflect the first two research
questions. Here is the analysis from Research Question 1 below the
hypothesis.
Research Question and Hypothesis 1
Research Question 1: What is the relationship between the number of
STEM courses taken in high school and postsecondary school by females
and their career choices?
Hypothesis One
Ho: The means of the number of STEM classes are the same for different
career choice categories
H1: At least one of the means of the number of STEM classes is not the same
for the different career choice categories
Hypothesis in Statistical Terms
Ho: 1=µ2=µ3=µ4=µ5= 6= 7
H1: At least one mean is different if I define categories as follows: H1 shows
that at least one mean is different 1≠µ2≠µ3≠µ4≠µ5
Since the seven groups were reduced to five, the hypothesis is
Ho: 1=µ2=µ3=µ4=µ5
H1: At least one mean is different if I define categories as follows: H1 shows
that at least one mean is different 1≠µ2≠µ3≠µ4≠µ5
1 to µ5 are now the factor groups reduced from the original seven. The
independent variable is the career choice categories. The five factor groups
are defined statistically below.
1= career choices for females in science/math (A)
2= career choices for females in technology/IT (B)
3= career choices for females in engineering (C)
4= career choices for females in nontechnical positions (D)
5= career choices for females in caring professions, humanities, and
education (E)
Research Question and Hypothesis 2
Research Question 2: What is the relationship between the number of
STEM role models in high school and postsecondary school and their career
choices?
Hypothesis Two
Ho: The number of female STEM role models in high school and
postsecondary school are the same for the different career choice categories
H1: The number of STEM role models in high school and postsecondary
school are not the same for the different career choice categories
Ho: 1=µ2=µ3=µ4=µ5
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different 1≠µ2≠µ3≠µ4≠µ5
The first set of data analyses in the first group of tables and figures
reflect the first two research questions and hypotheses stated above. The bar
chart in Figure 2 shows the total number of responses in each of the five
career choice categories: A science and math, B IT, D nontechnical
(business, legal, administration, etc), and E caring professions which include
some medical where some STEM classes may be required. There were no
respondents who chose engineering as a career choice so therefore, there
was no category C., therefore, it is excluded These categories correspond
with the numerical categories used in SPSS, indicated above with A as 1, B
as 2, D as 4, and E as 5.
The average numbers of respondents in the different career choice
categories is shown in the following bar chart. With the average, there is less
of a significant difference among the career categories with averages of 24
for A for science and math, 27, for B which is IT, 17 for D which is non
technical, and 27 for E which is caring professions.
Since the main purpose of the pilot was to determine the validity of
the survey instrument and there were only four respondents, I conducted the
data analysis for the pilot together with all the university responses. For the
survey university responses 43 surveyed of the general sample, since
nontechnical careers (D) were the broadest category, 50% of the respondents
were in these types of careers which include any non STEM education, any
soft science such as business or the social sciences or any administrative or
other professions. Like those who chose STEM (A) or SPSS category 1 or
IT (B) or SPSS category 2 careers, even those who chose nontechnical
careers (D) or SPSS category 4 did take more STEM classes and had more
STEM role models than those in caring professions with the exception of
nursing or medical (E) or SPSS category 5. Since no one in the sample chose
an engineering degree, there was no category (C) or SPSS category 3.
The reason it appears that the nontechnical group took more STEM
classes and had more STEM role models than the other groups was merely
because the number of females in nontechnical fields was larger in numbers.
Moreover high school and in the freshman year in college, certain math and
science classes are required. There were 33% of the respondents that had
chosen STEM careers (A) or SPSS category 1 and 11% that chose caring
professions (E) or SPSS category 5. Those who chose STEM careers and
some social science careers took more STEM courses in high school,
college, and graduate school and therefore tended to have more role models.
Table 1 is the ANOVA table for the survey respondents using a .05 p
value and 95% confidence interval with degrees of freedom of 3. The
resulting p value in this case was .000 possibly because even some of the
caring professions like nursing or medical take a large number of STEM
classes and have additional female role models as do females in
nontechnical professions like business or legal where math is required,
reducing the significance slightly (Field, 203). Therefore, there is a
significant difference between the number of STEM classes and the career
choices among the career choice groups. This first ANOVA was a general
one using both STEM and Role which is shown in the Appendix. Table 1,
however, shows the subset of the number of STEM classes versus career
choices across all five mean groups of career choices. Then to answer each
of the first research questions and determine those hypotheses, I conducted
separate
ANOVAs, one with career choices and STEM and the other with career
choices and role to determine a true significance. The first ANOVA table
shows the significance with career choices and both the number of STEM
courses and role models together. The second two show the individual
ANOVAs based on the first two research questions and hypotheses.
Table 1
One Way ANOVA: Number of STEM Classes RQ and Hypothesis 1
Source Sum of
squares
df Mean of
squares
F Significance
Intergroups 728.272 4 242.757 8.449 .000
Intragroups 1264.208 44 28.732
Total 1992.479 48
Table 1 was the One-Way ANOVA with STEM classes and career
choices. The F value is 8.449 with a total sum of squares for all responses of
1992.479. The mean of squares across the groups was 242.75 and within the
groups are 28.73. The significance is .0000 making the difference across
means very significant since the p value is below .05 (Field, 2013; Morrow,
2013; Nachmias & Nachmias, 2008). This means that the career choices a
female chose were significantly related to the number of STEM classes she
took in high school and postsecondary education especially for categories of
science and nontechnical.
The residual sum of squares demonstrates the error in the model in
prediction (Field, 2013). Since there is a large significant difference among
the five career choice categories in relation to the number of STEM classes,
I would reject the null hypothesis because the p value is.000 showing a
significant difference among the group means which are not the same across
means as the null hypothesis was indicative of (Dunteman & Ho (2006).
Moreover, according to Dumteman & Ho (2006), if F is larger than 1,
usually the null hypothesis is rejected and it is quite larger than 1. Since
there were no engineering responses only the other four categories had data,
which would make the degrees of freedom three instead of four.
Table 2 is the one way ANOVA for role models versus career choices.
In this table, the total sum of squares is 37.27, the degree of freedom
intergroup is 3 and the mean squares across groups are 1.78. Moreover the
significance is .08 which is slightly above .05 making the difference
between the number of role models and one’s career choices, less
significant. In other words, there is not a significant relationship with the
number of role models one had in school and whether or not the respondent
chose a career in math and science, IT, nontechnical, or a caring profession.
Table 2
One Way ANOVA: Number of Role Models RQ and Hypothesis 2
Sum of df Mean Squares F Signification
Squares
Intergroups 5.355 3 1.785 2.405 .080
Intragroups 31.921 43 .742
Total 37.277 46
For Table 2 above, the F ratio is 2.405, the degree of freedom among
the groups was three, and the p value was .08 making the number of role
models in relation to career choices, nonsignificant. This may be due to the
fact that some caring professions require science classes such as majors like
nursing and therefore, the student may have additional role models
encouraging her to take more science classes, for example.
As seen in the table it is necessary to use the sample size of the
bonded mean which is 5.249.The group bounded effect means are being
employed and the type one error levels are not guaranteed.
No respondent in this sample chose an engineering (C) career.
Moreover, the CI is 95%.The grand mean is 29.9773. This is a weighted
average. There were no engineering responses.
Table 3 is the multiple comparisons from the ANOVA among the
career choice groups and the significance between each career choice group
and the number of STEM classes and female STEM role models, still
reflecting research questions one and two. The only category that appears to
be significant in the relationship between career choices and the number of
STEM classes is the nontechnical category with a p value of .002. This may
be due to the fact that even business majors, and law majors have to take
science and math classes. There is also a significant difference between the
number of STEM classes and caring professions since some caring
professions may include nursing or healthcare which requires science and
math. There is no significance between the other categories especially
between math and science and IT where the p values are above .05.
For the role models, there does not seem to be a significant difference
between the career choices and the number of female STEM role models
across career groups.
Table 3
Multiple Comparisons: RQs and Hypotheses 1 and 2
LSD
Dependent (I) Career (J) Career Choice Mean Standard Significance CI 95%
Variable Choice Category Category Differences Error Lower
Upper
(I-J) Bound
Bound
IT -3.050 3.319 .363 -9.74 3.64
Math & Science Nontech 6.814* 1.656 .000 3.48 10.15
Caring Professions. -3.383 3.319 .314 -10.07 3.31
Math & Science 3.050 3.319 .363 -3.64 9.74
IT Nnontech 9.864* 3.299 .005 3.21 16.51
Number of
Caring Professions/ -.333 4.377 .940 -9.15 8.49
STEM
Math & Science -6.814* 1.656 .000 -10.15 -3.48
Classes
NonTech IT -9.864* 3.299 .005 -16.51 -3.21
Caring Professions -10.197* 3.299 .003 -16.85 -3.55
Math & Science 3.383 3.319 .314 -3.31 10.07
Caring
IT .333 4.377 .940 -8.49 9.15
Professions
Nontech 10.197* 3.299 .003 3.55 16.85
IT 1.100 .639 .092 -.19 2.39
Math & Science Nontech .555* .266 .043 .02 1.09
Caring Professions -.233 .533 .664 -1.31 .84
Math & Science -1.100 .639 .092 -2.39 .19
IT Nontech -.545 .636 .396 -1.83 .74
Number of Role Caring Professions -1.333 .787 .097 -2.92 .25
Models Math & Sciece -.555* .266 .043 -1.09 -.02
Nontech IT .545 .636 .396 -.74 1.83
Caring Professions -.788 .530 .145 -1.86 .28
Math & Science .233 .533 .664 -.84 1.31
Caring
IT 1.333 .787 .097 -.25 2.92
Professions
Nnontech .788 .530 .145 -.28 1.86
The next section is the data analyses for the linear regression which
reflects research question and hypothesis three.
Research Question and Hypothesis 3
Research Question 3: What is the relationship between salaries and the
number of STEM courses taken in high school and postsecondary school by
females, and the number of same sex role models?
Hypothesis Three
Ho: The salaries are independent of number of STEM courses in high school
and postsecondary school and/or role models.
H1: Salaries are dependent on the number of STEM courses in high school
and postsecondary school and/or role models.
H0: 1= 2=0, both betas are zero
H1: at least one Bs not equal 0
Y= o+ 1X1+ 2X2 where X1 is the number of STEM courses and X2 is the
number of same sex role models and Y=salaries
Linear Regression for Research Question and Hypothesis 3
Linear regression determines the relationship between the interval variables
by expressing the relationship as an algebraic equation by predicting
outcomes, according to Nachmias and Nachmias (2008). The residual sum
of the squares indicates how well the line fits the data, according to Field
(2013). In the equation Y= o+ 1X1+ 2X2 , Y was the dependent or outcome
variable of salaries, and X1 and X2 were the independent variables, or
predictor variables of the number of STEM classes and number of STEM
female role models. A multiple regression was conducted to determine the
relationship between the number of STEM classes taken and number of
same sex role models and salary, since this is a relationship study and not a
comparison between two means in research question 3 (Aczel, &
Sounderpandian, 2009; Field, 2013; Gibiliso, 2011; Nachmias & Nachmias,
2008). The midpoint of the salary range for each one was input into the Y
column of the first multiple regression chart and the total number of STEM
courses from high school to graduate school was input as X1 and the number
of female STEM role models from high school to graduate school was input
as X2.
The results of the regression were taken from input data of Y =
salaries, and X1 which is the number of STEM classes each individual took
in high school and postsecondary and X2 was the number of STEM female
role models. Although career choices was not a variable in the regression, I
still took it into consideration from what I learned in questions one and two
where the analysis of variance data revealed that the more role models and
STEM classes taken was directly related to STEM related career choices. I
used the same A for science and math, B for IT, C for engineering, D for
nontechnical, and E for education, caring, or humanities. Nontechnical can
also include social sciences and caring professions included nursing which
may dictate taking additional STEM classes and having such role models.
Moreover, no respondents chose engineering (C) careers. The trend
demonstrated from the linear regression was the linear relationship that
showed a direct but weak correlation with additional STEM classes and role
models with increased salaries, although the additional STEM classes had a
stronger correlation than role models. When I analyzed these data with that
of questions one and two, it seemed to correlate with those who chose
STEM careers as having higher salaries since these careers dictated taking
more STEM classes and having more STEM female role models.
The Linear regression showed that the more STEM classes and role
models a female had, she tended to earn a higher salary which also
correlated with choosing a STEM career in science or math (A) or IT (B)
based on the ANOVA from questions one and two. However, the
relationship was not a strong one because some careers like nontechnical (D)
that include business and legal require a great deal of IT and math courses
and some caring professions (E) such as nursing and medical, also require a
great deal of math and science.
In the following tables and figures, the independent variables were the
number of STEM classes and female STEM role models in high school and
postsecondary education and the dependent variables was salaries. There
were no engineering responses; therefore, there is no data in that category.
Table 4 is the descriptive statistics of the linear regression where the
dependent variable is salary and the independent variables are the number of
STEM classes and the number of STEM female role models. The mean is
51.06 for salary in 1000s, 21.02 for STEM classes and .81 for role models
since there are more STEM classes taken and fewer role models.
Table 4
Descriptive Statistics
Mean Standard N
Deviation
Salary in 1000s 51.06 33.312 48
Number of STEM Classes 21.02 6.418 48 Number of
Role Models .81 .900 48
Table 5 below is the sum of squares and the residuals and the mean of
squares. The significance between the salaries earned by the sampled female
alumnae and the number of STEM classes and the number of female role
models in secondary and postsecondary school is .337 (p = 0.33). This
means the relationship between the numbers of STEM classes females took
in this sample along with the number of female role models was not
significantly different than the salary they earned. With the p value of, 337,
this is not significant.
Table 5
ANOVAa
Model Sum of
Squares
df Mean of
Squares
D Sig.
Régression
1 Résiduals
2459.934
48586.875
2
44
1229.967
1104.247
1.114
.337b
Total 51046.809 46
Note. a. Dependent Variable: Salary in 1000s b. Values constant an
predicted, Number of Role Models, Number of STEM Classes, p value is
not significant, role models not significant to career choices
In Table 6 there are the linear correlations between salary and the
number of STEM classes and role models. In general, there is a positive
relationship with those who took more STEM classes and higher salaries.
However, because some caring professions include nursing and some
nontechnical include finance, accounting, and legal professions which tend
to pay high salaries, the relationship between the number of STEM classes
and STEM role models has a slightly weaker correlation using the Pearson
Correlation method. The correlation appears to be more significant for the
number of STEM classes and salary, then the number of role models and
salary. Therefore, the only significant relationship shown by .002 is the
relationship between the number of STEM classes with role models and not
as much with salary except for the number of STEM classes.
Table 6
Correlations
Salary in
1000S
Number
of STEM
Classes
Number
of
Role
Model
s
Pearson Correlation
Salary in 1000s
Number of STEM
Classes
1.000
.220
.220
1.000
.094
.418
Number of Role
Models
.094 .418 1.000
Sig. (unilateral)
Salary in 1000s
Number of STEM
Classes
.
.069
.069
.
.265
.002
Number of Role
Models
.265 .002 .
N
Salary in 1000s
Number of STEM
Classes
48
48
48
48
48
48
Number of Role
Models
48 48 48
The correlation and p value are given in Table 6 above where it
indicates sig (unilateral). The significance value in Table 6 between STEM
classes and salary is more significant at a p value of .002. The significance
value between role models and salary has a less significant relationship and
the p value is .265.
Table 7
Coefficient of the Correlation Results
Model R R R-Squared standard error of
Squared Adjusted estimation
1 .220 a .048 .005 33.230
Note. a. Values Number of Role Models, Number of STEM Classes
Table 7 above shows the R squared of the correlation which is .220
for the number of role models and STEM classes, the R squared at .048, and
when the R squared is adjusted it became .005. The R squared is the the
square of the correlation between the dependent (salaries) and independent
variables (STEM classes and role models). The adjusted R squared is less
biased. Only 5% of the variation in your dependent variable is explained by
your independent variable. The standard error of estimation is 33.23.
Histogram with salary as the dependent variable is 1000s with a mean
of 51.67 and a standard deviation of 33.22 for a sample size of 48. This
shows a skewed distribution as opposed to a normal one. The vertical is the
effective residuals and the horizontal scale is salary in 1000s. This histogram
is a graph of the residual behavior of salaries to determine the kind of
distribution, showing the impact of the residual sum of squares (Field,
2013). The residuals are there to determine if the histogram is centered in
distribution.
Scatter plot
Scatter plot including all 48 responses-STEM vs. Salary. This scatter
plot included all 48 responses including the pilot. Therefore, there was no
valid result of the Collinearity and correlation of the variables with the pilot.
The scatter plot reflects the multiple linear regression model of the data
collected by the university respondents with the outcome Y variable as
salaries, and the two predictor variables are X1 STEM classes taken and X2
STEM female role models. The first scatter plot is the relationship between
STEM classes and salary which tends to be a positive relationship, the more
STEM classes one takes, the higher their salary since science, math, and IT
fields tend to offer higher salaries. The Pearson Correlation Coefficient is
used to measure the strength of the relationship between the two variables. R
is 0.4967, with a p value of .002 for STEM classes and .265 for salaries,
showing a weak relationship between the variables, according to the scatter
plot, but the p value of .002 showed a significant relationship. The Pearson
Coefficient for the independent variables was .069.
The results from the scatter plots were from the input of the original
data from all of the university respondents. The scatter plot shows a weak
relationship between the number of STEM classes and salaries, but there is a
positive relationship with the p value showing a more significant one.
Moreover, this relationship is more significant than Role Models and Salary.
The relationship between salary and the number of STEM cases in
Figure 4 is clustered in the lower center of the graph, which demonstrates a
partial regression. The salary is in 1000s and therefore, is somewhat of a
positive correlation between the numbers of STEM classes which increase,
resulting in higher salaries.
Partial regression between salaries and role models, using 48 data
points. The Pearson Correlation Coefficient is used to measure the strength
of the relationship between the two variables, which here show no
relationship. R is 0.9171 which is confusing because that would indicate a
strong relationship and yet the scatter plot shows no relationship between the
salary and the number of STEM female role models. There is more
information above in Table 6 for the Pearson Coefficient, which was .069
between the independent variables.
Descriptive Analysis of Demographics
In analyzing the demographics, again, there was a total of 4 pilot and
40 respondents who were all female alumnae in from bachelor’s or master’s
programs in Long Island universities or the Walden participation pool. Each
respondent was a female born after 1980, from all income backgrounds from
individual salaries of under $10,000 to $120,000. Many of the salaries seem
to concentrate between $40,000 and $60,000.
Analysis of Independent Variables
For questions one and two, for the ANOVA, the independent variable
was career choices where it was analyzed that when females chose STEM
career choices such as science and math (A) or IT (B) they tended to take
more STEM classes and have 1 to 3 STEM role models. The independent
variables for the multiple linear regression in Research Question 3 were the
number of STEM classes and female STEM role models in high school and
postsecondary school which corresponded with Research Question 1 where
the higher salaries which was the dependent variable correlated with taking
more STEM classes and having more role models and choosing STEM
careers.
Analysis of Dependent Variables
The dependent variable in questions one was the number of STEM
classes taken in postsecondary education which corresponded with the
independent variable of career choices. Females who tended to choose
STEM careers tended to take more STEM classes. The same is true in
Research Question 2 for STEM female role models. Those who chose
STEM careers and took more STEM classes tended to have more female
STEM role models as they appeared to be encouraged to take STEM classes
and choose such careers.
Descriptive Statistics--Residual Plots and Scatter Plots
For the residuals, which are the errors, in this residual plot are
assumed to have multiple multicollinearity, with the variance inflation factor
and tolerance. To identify multicollinearity is to view the matrix for the
degree of correlation (Field, 2013). The dependent variable is salaries and
the independent variables are the numbers of STEM classes taken and
STEM female role models in high school and postsecondary education.
These residuals represent errors and the degree of correlation represents a
positive correlation between the number of STEM classes and the number of
STEM female role models and how they correlate with salaries.
In order to determine the correlation between the numbers of STEM
classes and STEM role models in high school and postsecondary school and
their positive relations with salaries, which tend to be higher when choosing
STEM science/math (A or 1) careers or IT (B or 2) careers, are shown here.
Sometimes females who took non tech social science (D or 4) careers may
have taken a lot of STEM classes. Subsequently they may have had STEM
role models. Furthermore, some females may have taken caring professions
(E or 5) and the reason they may have taken additional STEM classes and
had increased numbers of female role models may be because these
professions also include nursing and medical which require additional
science courses. There were no respondents who had engineering careers,
coded by Cor 3).
Sample Selection
Randomness increases validity and reliability of the sample by
insuring equal chance of participation. However, because the specific
population was female alumnae from the chosen sampled universities born
on or after 1980, I employed a stratified sample. Hence, this was a stratified
simple random sample (SRS) (Kalton, 1983). Therefore, bias is reduced. The
rationale for using a random sample or a systematic sample with a random
starting point was that each person has an equal chance of participation in
the study, which reduces selection bias (Case, 2007; Field, 2013; Nachmias
& Nachmias, 2008, Patton, 2009). The sample will was randomly drawn by
each alumni association at each sample university to randomly extract the
female alumni born after 1980 for this study, using the alumni association as
gatekeepers.
The process to obtain the sampling frame from each of the alumni
from the four universities alumni centre was conducted by providing the link
to the alumni gatekeeper who provides a link to alumnae randomly in
choosing which female alumnae will be studied (Kalton, 1983). The
stratification is geographic taking place in Long Island. Moreover, since the
sample size was drastically reduced, I expanded it slightly by making the
survey also available in the Walden pool of participants. Since I had no
contact with the respondents, this ensured their privacy and confidentiality
clause (Kalton, 1983; NIH 2008). Before anyone was contacted, I obtained
IRB approval with the reference number 11-18-15-0169928 (NIH, 2008). To
perform the lottery method, the alumni association forwarded the survey link
to the alumnae randomly. I had no contact with the respondents ensuring
anonymity.
In this statistical construct using an ANOVA, in these factor groups,
the dependent variable is the number of STEM classes and the independent
variable is the career choices. Using a one way ANOVA, I determined that
the average numbers of STEM classes taken are different across factor
groups which are career choice categories.
The trend was that females who chose STEM career categories tended to
take more than STEM classes than those who do not choose such career
categories. I tested to see if career choice categories are related to the
number of STEM course taken in the past and using other tests I compared
some of the categories to see if they are the same statistically. Those who
chose STEM careers took more STEM classes. The grand mean average is
42.25. This is a weighted average to attach the hypothesis.
The math and science categories were combined in M1 (A) and IT
was M2 (B). Photonics and research and development are included in
engineering (C). There were no respondents who chose a career in
engineering. Caring professions (E) are healthcare, nursing, medical, and
home health aides. Since nursing is included, there were some who chose
these professions who took more STEM classes. Education is teachers,
professors, or anyone who works in a school district or postsecondary
institution, which were included in the same M5 (E) category as the caring
professions. Nontechnical (D) includes those professions that are not in a
STEM, caring, or educational profession (including business, administrative,
service, retail, manufacturing, and legal). Since business requires a large
number of math courses, some females who chose these careers did take
more math classes.
In using ANOVA, is a procedure to test the hypothesis in order to
evaluate the differences in the means among the seven groups below
(Iverson & Norpoth, 1987; Morrow, n.d.). When I analyze the data, that
there are too few responses in one of the seven categories, it was necessary
for me to combine categories down to 5. I combined math and science, and
caring with education and humanities.
Results
Summary of Hypothesis 1
Research Question 1: What is the relationship between the number of
STEM courses taken in high school and postsecondary school by females
and their career choices?
Hypothesis One
Ho: The means of the number of STEM classes are the same for different
career choice categories
H1: At least one of the means of the number of STEM classes is not the
same for the different career choice categories Hypothesis in statistical
terms:
Ho: 1=µ2=µ 3=µ4=µ5=µ6=µ7
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different 1≠ µ2≠µ3≠µ4≠µ5
Since the 7 groups were reduced to 5, the hypothesis is:
Ho: 1=µ2=µ3=µ4=µ 5
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different 1≠µ2≠µ3≠µ4≠µ5 µ 1 to µ 5 are now the factor
groups reduced from the original 7. The independent variable is the career
choice categories. The 5 factor groups are defined statistically below.
µ 1 = career choices for females in science/math (A) µ 2 = career choices
for females in technology/IT (B) µ 3 = career choices for females in
engineering (C) µ 4 = career choices for females in nontechnical
positions (D) µ 5 = career choices for females in caring professions,
humanities, and education (E)
The hypothesis is that there is a positive relationship between the
numbers of STEM classes taken in high school and postsecondary school
and choosing a STEM career. I used the LSD post-hoc test to identify which
courses have the highest significance, which tended to be the STEM classes
and STEM careers. This is important because ANOVA does not tell which
of the categories are different, only that at least two of the categories are
different. Post hoc tests determine the greatest differences (Field, 2013).
Based on the sample evidence, using the ANOVA, the idea that the
relationships are equal across factor groups was rejected, but there was not a
significant difference between the variables.
There is a difference between the means with dispersion around the
respective means. This measures how the observations differ of these group
means.
The 7 original factor groups have been used in Research Question 1
have been reduced to the same 5 used in Research Question 1. These factor
groups as well as the statistical analysis will be the same for Research
Question 2 as was indicated in Research
Question 1. The only difference is that the dependent variable is the number
of female STEM role models instead of the number of STEM classes taken.
The above discussion can be summarized by the ANOVA Table 1 for
Research
Question and Hypothesis 1 on page 173 and 174.
The sum of squares SS inter-groups between the group means and the
grand mean quantifies the variability between the groups of interest of
728.272, The SS (Error) is the sum of squares between the data and the
group means which quantifies the variability within the groups of STEM
classes of 1264.208. The Total is the sum of squares is 1992.479.
The mean squares (MS) are the average sum of squares for the factor
and the error: The F column contains the F-statistic which is the average
variability within the groups, the ratio of the Between Mean Sum of Squares
to the Error Mean Sum of Squares. The F statistic indicates if the means in
an ANOVA are significantly different and helps to determine the p value.
With a p value of .000 and F ratio of 8.449 the number of STEM classes in
regard to career choices is significantly different therefore the null
hypothesis is rejected.
Summary of Hypothesis 2
Research Question 2: What is the relationship between the number of
STEM role models in high school and postsecondary school and their career
choices?
Hypothesis Two
Ho: The number of female STEM role models in high school and
postsecondary school are the same for the different career choice categories
H1: The number of STEM role models in high school and postsecondary
school are not the same for the different career choice categories
Ho: µ1=µ2=µ3=µ4=µ5
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different µ1≠µ2≠µ3≠µ4≠µ5
In this statistical construct using ANOVA, in these same factor
groups, the dependent variable is the number of same-sex STEM role
models and the independent variable is the career choice categories. Using a
one way ANOVA, I determined if the average numbers of STEM same-sex
role models are different across factor groups which are career choice
categories. The hypothesis is that there is a positive relationship between the
numbers of STEM classes taken in high school and postsecondary school
and choosing a STEM career (Farland-Smith, 2009). Under the null
hypothesis, using an ANOVA, the relationships are equal across factor
groups. The more same-sex STEM role models a female has, the more likely
she is to choose a STEM career. The 5 factor groups are the same for both
questions one and two. Again, similar to Research Question 1, my intention
is to retrospectively investigate the number of STEM same-sex role models
are different by career choice categories which are the factors. I tested to see
if career choice categories are related to the number of same-sex STEM role
models females had in the past. Using pairwise multiple comparisons and
the ANOVA, there is a weak relationship between the variables of role
models and career choices. The p value for the pilot is .05 and for the rest of
the responses for same-sex role models is .08. This demonstrates a
relationship between the variables but it appears that the pilot showed a
more significant relationship between choosing a STEM career and the
number of STEM classes taken and female role models than the rest of the
responses. STEM classes and career choices had a more significant
relationship than role models as indicated in Table 1 on page 174. While
there is still a slightly weaker relationship, being the p value is .08, it seemed
that the number of role models was not as significantly related to the career
choices as STEM classes because some nontechnical alumnae also took a lot
of STEM classes and had increased STEM same-sex role models. An
increase in the F-value can decrease the pvalue; increasing the significance
and it is significant if below .05 and less significant the more above .05 it is
(Field, 2013; Morrow, 2013; Nachmias & Nachmias, 2008).
Moreover, some in nontechnical professions may have taken a lot of
STEM classes and had same-sex STEM role models particularly in business
or accounting where there is a need for a lot of math or IT classes, or social
sciences who take additional science classes. Even some caring professions
required some additional STEM classes such as nursing which skewed the
results and lessened the significance between choosing a STEM career with
the number of STEM classes and STEM role models. Although the
relationship showed a slight decrease in significance, there is a pairwise
comparison of group means demonstrating there is still a relationship,
resulting in accepting the alternative hypothesis and rejecting the null
hypothesis.
This above discussion can be summarized by Table 2 which is on page 174.
See Hypothesis 1 for meanings of SS, df, MS, and the F statistic. The
meanings are the same for both Hypotheses 1 and 2. For role models, the
relationship is less significant with a p value of .08 and F statistic of 2.405.
There is still some difference so the null hypothesis is rejected, but less
significant relationship, therefore role models is less indicative of career
choices.
Summary of Hypothesis 3
Research Question 3: What is the relationship between salaries and
the number of STEM courses taken in high school and postsecondary school
by females, and the number of same sex role models?
Hypothesis Three
Ho: The salaries are independent of number of STEM courses in high school
and postsecondary school and/or role models.
H1: Salaries are dependent on the number of STEM courses in high school
and postsecondary school and/or role models.
H0: 1= 2=0, both betas are zero
H1: at least one Bs not equal 0
Y= o+ 1X1+ 2X2 where X1 is the number of STEM courses and X2 is the
number of same sex role models and Y=salaries
The number of STEM courses and the number of same sex role
models are the independent or predictor variables and salaries is the
dependent or outcome variable. This is a multiple regression. In this research
question, I established a linear relationship between the independent and
dependent variables (Field, 2013). The hypothesis is that
the number of STEM classes taken and the number of same-sex STEM role
models have a positive relationship with a higher salary since if these
conditions exist, it is more likely females will choose STEM careers which
tend to have higher salaries. With an .05 alpha means I have only 5% chance
of being wrong when Ho is true and is rejected, about the relationship
between the independent and dependent variables about the relationship
between participation in math and science classes through the number of
classes females took, the number of female role models and their impact on
career choices for females facing barriers to entering these STEM fields.
Thus, the confidence interval is 95% which means there is a 95% likelihood
that the interval contains the true limits where the population mean is likely
to fall (McAllister, 2015). The confidence interval is known parameter of
participation in math and science correlate with choosing a STEM career
success with the ability to earn a higher salary than without choosing a
STEM career (Burkolder, 2010; Field, 2013; Gibilisco, 2011; Green &
Salkind, 2011; Nachmias & Nachmias, 2008). Some females may choose a
nontechnical or caring career and still take a lot of STEM classes because
they are either required or the career has a great deal of math like finance or
business or a lot of science like nursing. It is the proportion of variance in
my dependent variable of salaries that is accounted for by my set of
independent variables (Field, 2013; Morrow, 2013).
The salaries ranged from $10,000 to $120,000. The alumnae who
earned $10,000 or $20,000 in salaries were generally graduate students who
were alumnae of bachelor’s programs or they were employed in caring (E)
professions. The alumnae who earned $60,000 or more were generally in
science or math (A) or IT (B) while some were in nontechnical professions
and a few caring professions. There were some alumnae who earned
$40,000 who were in the science and math professions. The trend is that the
higher salaries tend to have taken more STEM classes and have more STEM
female role models which tend to be in science and math (A) or IT (B).
However, the relationship between salary and role models was considerably
less significant than those between salary and STEM classes. There were no
engineering respondents (C). Salaries in the A and B categories were as high
as $120,000. The p value after the regression was .04 which still
demonstrated a significant relationship among the variables allowing the
ability to accept the alternative hypothesis and reject the null hypothesis.
There was also significance with the residuals of .337 with the relationship
between the salary earned and the number of STEM classes and female role
models a female had in secondary and postsecondary education being
insignificant. In other words the number of STEM classes and role models
are not significant determinants of salary.
Outliers can affect the precision of the estimation of the regression
weights, making data cleaning necessary before the regression is
conducted. The data cleaning was conducted by deleting any outliers
(Morrow, 2011). Therefore, there were not any outliers. As seen from the
histogram, the distribution was close to normal. There was little linearity
as evident by the weak relationship shown in the scatter plots. In order to
achieve multicollinearity using the Pearson's Bivariate Correlation among
all independent variables the correlation coefficients, need to be smaller
than .08, which for the independent variables was .069. Autocorrelation
occurs when the residuals are not independent from each other, which was
not the case here. There is no Homoscelasticity as evident from the
irregular weak correlation evident in the scatter plots. The above
discussion can be summarized by the correlation Table 6 on page 182.
The correlation and p value are given in Table 6 above where it
indicates sig (unilateral). The significance value in Table 6 between STEM
classes and salary is more significant at a p value of .002. For this, the null
hypothesis is rejected. The significance value between role models and
salary has a less significant relationship than the significance between
STEM classes and salary with the p value is .265. The relationship is weak
with little significance but there are still some minor differences which is
why the null hypothesis is still rejected.
Here is also the descriptive statistics for the hypothesis of Research
Question 3 with the independent variables of the numbers of STEM classes
and role models and the dependent variable of salaries, which is the same as
Table 4 on page 181. The more STEM classes a female took, generally the
higher the salary. This relationship was more significant than the
relationship between the number of role models and salaries as seen from the
p value of .265. The mean salary is 51,006 dollars with a standard deviation
of 33.312 for a sample of 48. The mean and standard deviation for the
number of STEM classes is 21.02 and 6.4, respectively. The mean and
standard deviation for the number of role models is .81 and .9, respectively.
See Table 4 from page 181, Descriptive Statistics for Research Question and
Hypothesis 3.
Chapter Summary
In this chapter, I described the data collection, analysis, and results of
the study. In answering the first two research questions by using the
ANOVA, The pilot was primarily designed to check and improve the survey
and to determine any changes necessary to the instrument. The results were
combined since this is a small data set. The respondents had a less
significant relationship with a p value of .07 but still somewhat significant,
therefore, I still was able to reject the null hypothesis and accept the
alternative hypothesis. The limitation here was that the sample size was
considerably smaller than I had originally forecasted. However, the sample
was randomly extracted from the alumni associations of four universities
which increased the validity and reliability of the responses through
randomness of a simple stratified random sample. I have analyzed the data
collected on each of the three research questions where the data were shown
in tables, figures, histograms, and scatter plots, demonstrating a significant
relationship between the number of STEM classes with salaries and career
choices with a weaker relationship between these variables and role models.
This chapter also summarized the hypotheses which are the following:
The summary of hypothesis one from Research Question 1 is that
there is a positive relationship between the numbers of STEM classes taken
in high school and postsecondary school and choosing a STEM career. I
used the LSD post-hoc test to identify which courses have the highest
significance, which tended to be the STEM classes and STEM careers. This
is important because ANOVA does not tell which of the categories are
different, only that at least two of the categories are different. Post hoc tests
determine the greatest differences (Field, 2013). Under the null hypothesis,
using the ANOVA, the relationships are equal across factor groups, which
was rejected and the hypothesis was accepted.
The summary of hypothesis 2 for research question 2, using a one way
ANOVA, I have determined if the average numbers of STEM same-sex role
models are different across factor groups which are career choice categories,
and then I can accept the alternative hypothesis which there was a
categorical difference, unlike under the null hypothesis. The hypothesis is
that there is a positive relationship between the numbers of STEM classes
taken in high school and postsecondary school and choosing a STEM career
(Farland-Smith, 2009). Under the null hypothesis, using an ANOVA, the
relationships are equal across factor groups. The more same-sex STEM role
models a female has, the more likely she is to choose a STEM career.
The summary of hypothesis 3 and research question 3 is the trend is
that the higher salaries tend to have taken more STEM classes and have
more STEM female role models which tend to be in science and math or IT.
However, the relationship between salary and role models was considerably
less significant than those between salary and STEM classes. There were no
engineering respondents. Salaries in the A and B categories were as high as
$120,000. The p value after the regression was .04 which still demonstrated
a significant relationship among the variables allowing the ability to accept
the alternative hypothesis and reject the null hypothesis. There was also
significance with the residuals of .337 with the relationship between the
salary earned and the number of STEM classes and female role models a
female had in secondary and postsecondary education being insignificant. In
other words the number of STEM classes and role models are not significant
determinants of salary.
Chapter 5: Discussion, Conclusions, and Recommendations
Summary
In summarizing this study, the objective was to determine the strength
of the relationship between the number of STEM classes females born after
1980 from the sampled school alumnae took in high school and
postsecondary education with the the number of STEM female role models
during this same period with their career choices and salaries using a survey
instrument with a 5-point Likert scale. The data analysis employed was an
ANOVA with the comparison across means of the career choice categories
and their relationship with the STEM classes and role models. With an
ANOVA to test the significance of the relationship and to test the
hypothesis, it was necessary to compare the groups of means and determine
if any are different.
The second method was the linear regression, which used an ANOVA
and regression to determine the relationship between salary and the number
of STEM classes and role models in high school and postsecondary
education. The conclusion shows a more significant relationship between
career choices and the number of STEM classes than between career choices
and role models. There was a less significance between the number of
STEM female role models and career choices. The results also showed a less
significant relationship between salary and the number of STEM classes and
role models. Moreover, in this chapter, I discuss the significance of the
study from Chapter 1 as well as the data analysis for the research questions,
limitations, and delimitations of the study from Chapter 1.
Conclusion
In concluding this study, it appeared that when I conducted the
ANOVA for the relationship between the number of STEM classes and
career choices was more significant, than the relationship between role
models and career choices. The relationship between math and science and
nontechnical appeared more significant with career choices. There was also
some significance with caring professions, probably because some of these
professions include nursing or healthcare, which requires science and math
courses. The linear regression showed there was less of a significant
relationship between salary and the number of STEM classes and role
models. In this study, the result was a small sample of 48 data points where I
had to draw conclusions from a small data set, substantially reduced from
the sample originally proposed in chapter 3. For this reason, the significance
showed a weak relationship among the variables and role models and
increased deviations and residuals. The relationship between STEM classes
and career choices and salaries was more significant than role models and
career choices and salaries. There is need for further study, possibly with a
broader geographic location and a larger sample size, to determine the
relationship among the variables with less residuals and increased
significance. In other words, the more science classes one took specifically,
the slightly higher number of role models. The most significant relationship
is between the number of STEM classes and career choices. The relationship
between the number of STEM classes and the number of female role models
versus salaries was weaker than the relationship between these variables and
career choices. However, the relationship between STEM classes and
salaries was slightly stronger for science and math and nontechnical careers
such as law and business, which tend to take a lot of math courses.
It is also hoped that this and other studies like it will impact social
change by reducing the gender gap in STEM classes and careers and that the
number of female role models in STEM will increase for young females in
the future. In the future, a t test can also be conducted with career choice
categories A or 1 science and math and D or 4, nontechnical because
nontechnical fields include business, legal, and social sciences, and many
take a considerable number of STEM classes, particularly math, science, and
IT. Subsequently, there are also several caring professions who take a
considerable number of STEM classes if the career is nursing, physician
assistance, or healthcare. Moreover, it is hoped to be able to publish the
dissertation findings to bring about the implication of social change. It is
also hoped that stereotypes that claim that females are not as proficient in
math and science as males will be dispelled and these fields will no longer
be associated with masculinity.
Significance of Study
As stated in chapter 1, this study is significant to society at large
because it may increase the understanding as to why there is a gender gap in
the STEM fields and how to close this gap through education (Carrell et al.,
2010). Dispelling such preconceived notions that females are not as good in
math or the issue of the lack of female role models in STEM fields may be
addressed, thereby helping females to increase their access to these higher
paying careers. Furthermore, according to Eccles and Wang (2016), females
tend to prefer language and humanities over math as they advance into
adolescence, and although they obtain higher grades than their male
counterparts, the males score better on standardized high stake exams in
math. Although females are well represented in healthcare and medical
fields (Eccles & Wang, 2016), they are still underrepresented in engineering
and other STEM fields, as evident from the fact that no female in my study
chose an engineering career. The lack of female interest in engineering
degrees was evident in my small sample, and according to Bystydzienski et
al. (2015) may contribute to the underrepresentation of females in this field.
Moreover, it is important to denounce Acker’s (1990) masculinity
theory where he postulated that females who work in STEM fields that are
traditionally masculine are out of their natural element. Females have similar
natural abilities in STEM to males (Dugan et al, 2013; Farland-Smith, 2009).
The results of this research can reveal to females how important science and
math are early in life and how parents should encourage their daughters to
be interested in math and science as children. By encouraging little girls to
explore math and science as children and through seeing female characters
portrayed in STEM careers such as the character of Dr. McStuffing can
entice young girls toward STEM at a young age. Moreover, it is hoped that
this type of research will further encourage young girls to explore STEM
careers and interests through these female character role models. Even
today, according to Bottia, Stearns, Mickelson, Moller, and Valentino
(2015), the underrepresentation of females in STEM from high school to
postsecondary, especially in areas like physics (Riegle-Crumb & Moore,
(2014), is a serious issue given the social and economic inequities that result
for females by not having the same access to these careers, even though my
study may not have shown the significant differences it intended to because
of the small data set.
By generating quantitative data on the relationship between the
numbers of STEM classes females take and the number of female role
models and the impact on career choices and salaries available, this
information might help females to better manage their course selections in to
be competitive in their career choices within the STEM field. These data
might also facilitate guidance counselors and deans to aid females on
counseling on how to better manage STEM careers, both academically and
in the workplace in this broad science of management.
Data Analysis for Research Questions
For the first two research questions, I employed an ANOVA whereby
comparing the means of five groups of career choices as the independent
variable and the number of STEM classes as the dependent variable for the
first question and the number of STEM role models as the dependent
variable for the second question. The five groups were reduced from the
original seven groups discussed in Chapters 1 and 3. The five groups are as
follows:
µ1 = career choices for females in science/math (A) µ2 = career choices
for females in technology/IT (B) µ3 = career choices for females in
engineering (C) µ4 = career choices for females in nontechnical
positions (D) µ5 = career choices for females in caring professions,
humanities, and education (E) The nontechnical positions include
business, legal, and administration, which were in category D. The E
caring professions included some medical where some STEM classes
may be required. There were no respondents who chose engineering as
a career choice so, therefore, there was no category C. These categories
correspond with the numerical categories used in SPSS, indicated
above with A as 1, B as 2, D as 4, and E as 5. There were no
engineering responses, and although no female in the sample chose an
engineering career, Bystydzienski et al. (2015) still emphasized the
importance of females having access to engineering degrees. The result
of my study demonstrates the urgency of this issue.
Response to Research Questions
Research Question and Hypothesis 1
Research Question 1: What is the relationship between the number of
STEM courses taken in high school and postsecondary school by females
and their career choices?
Hypothesis One
Ho: The means of the number of STEM classes are the same for different
career choice categories
H1: At least one of the means of the number of STEM classes is not the same
for the different career choice categories
Hypothesis in statistical terms
Ho: 1=µ2=µ3=µ4=µ5= 6= 7
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different µ1≠µ2≠µ3≠ µ4≠µ5 6≠ 7
Since the 7 groups were reduced to 5, the hypothesis is:
Ho: µ1=µ2=µ3=µ4=µ5
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different µ1≠µ2≠µ3≠µ4≠µ5
Relationship Between Career Choices and STEM Classes
In response to the first question, the relationship between career
choices and STEM classes was a positive one, with more significance than
career choices and role models. However, the research still showed this
relationship demonstrating Erikson’s (1980) theory. For example, one
woman said that when she was little, she enjoyed playing with dolls and
caring roles and she did not take many STEM classes; she chose a
nontechnical career, which also seemed to synchronize with Erikson’s
theory that what occurs early in life impacts what happens later in life.
Moreover, the fact that no female in the sample chose an engineering career
aligned with Acker’s (1990) theory about females feeling unnatural in
traditionally male occupations like engineering, which needs to change.
Research Question and Hypothesis 2
Research Question 2: What is the relationship between the number of
STEM role models in high school and postsecondary school and their career
choices? Hypothesis Two
Ho: The number of female STEM role models in high school and
postsecondary school are the same for the different career choice categories
H1: The number of STEM role models in high school and postsecondary
school are not the same for the different career choice categories
Ho: µ1=µ2=µ3=µ4=µ5
H1: At least one mean is different-If I define categories as follows, H1 shows
that at least one mean is different µ1≠µ2≠µ3≠µ4≠µ5
For the first two research questions, no one in the sample chose an
engineering career, took any engineering courses, or had any role models in
the field. The largest category was science, which includes all the hard
sciences such as earth science, biology, chemistry, physics, and astronomy.
According to Riegle-Crumb and Moore, (2014), there has traditionally been
fewer females than males in high school physics. The second largest
category was nontechnical, and the lowest categories, other than
engineering, were IT with only three responses. Nontechnical encompassed
everything from legal to administrative to soft sciences like business,
economics, sociology, or political science, which are professions that many
of the females in the study chose. Caring professions included social
workers, home health aides, nursing and medical; the latter two overlapped
with science and math since the STEM course requirements for both are
similar.
Since nursing was included as a caring profession (E), those who have
taken nursing classes have taken a larger number of STEM classes than
those in any other caring profession since nursing is a scientific field that
crosses over into the caring professions, making it an exception to the
hypothesis that those in caring professions have taken fewer STEM classes
than those in STEM fields. Also, females nontechnical fields that included
soft sciences may take some STEM classes. There were significant
differences between A, which were the hard sciences, and D, which were the
nontechnical, including the soft sciences; therefore, further research can be
conducted using a t test. Thus, the alternative hypothesis is accepted, since
this relationship was the most significant.
Relationship Between Career Choices and Role Models
The relationship between career choices and the number of female
STEM role models was weaker and less significant than the relationship
between career choices and the number of STEM classes. Similar to those
who chose STEM (A), which was also SPSS category 1 or IT (B), which
was SPSS category 2 careers, the respondents in the sample who chose
nontechnical careers (D) which was SPSS category 4 did take more STEM
classes and had more STEM role models than those in caring professions,
with the exception of nursing or medical (E) which was SPSS category 5.
Moreover, the reason it appeared that the nontechnical group took
more STEM classes and had more STEM role models than the other groups
was because the number of females in nontechnical fields was larger than
the other career choice fields. Moreover, in high school and in the freshman
year in college, certain math and science classes are required. There were
33% of the respondents who had chosen STEM careers (A) or SPSS
category 1 and 11% who chose caring professions (E) or SPSS category 5.
Those who chose STEM careers and some social science careers took more
STEM courses in high school, college, and graduate school and therefore
tended to have more role models. However, the difference was very
insignificant. For example, a person who chose a STEM career over
someone who chose a nontechnical or caring profession may have one more
role model in their entire career. Therefore, the relationship between the
career choices and the number of STEM role models was positive but quite
insignificant and weak. Therefore, the alternative hypothesis is only
marginally accepted for RQ 2.
Research Question and Hypothesis 3
Research Question 3: What is the relationship between salaries and the
number of STEM courses taken in high school and postsecondary school by
females, and the number of same sex role models?
Hypothesis Three
Ho: The salaries are independent of number of STEM courses in high school
and postsecondary school and/or role models.
H1: Salaries are dependent on the number of STEM courses in high school
and postsecondary school and/or role models.
H0: 1= 2=0, both betas are zero
H1: at least one Bs not equal 0
Y= o+ 1X1+ 2X2 where X1 is the number of STEM courses and X2 is the
number of same sex role models and Y=salaries
Relationship Between the Number of STEM Classes and Number of Role
Models
Versus Salaries
In the equation Y = o+ 1X1+ 2X2 , Y was the dependent or outcome
variable of salaries, and X1 and X2 were the independent variables, or
predictor variables of the number of STEM classes and number of STEM
female role models. A multiple regression was conducted to determine the
relationship between the number of STEM classes taken and number of
same sex role models and salary, since this was a relationship study and not
a comparison between two means in research question 3 (Aczel, &
Sounderpandian, 2009; Field, 2013; Gibiliso, 2011; Nachmias & Nachmias,
2008).
The results of the regression were taken from input data of Y =
salaries, and X1 which is the number of STEM classes each individual took
in high school and postsecondary and X2 was the number of STEM female
role models. Although career choices was not a variable in the regression, I
still took it into consideration from what I learned in questions one and two
where the analysis of variance data revealed that the more role models and
STEM classes taken was directly related to STEM related career choices. I
used the same categories I used for the first two research questions which
are: A for science and math, B for IT, C for engineering, and D for
nontechnical and E for education, caring, or humanities. Nontechnical can
also include social sciences and caring professions can include nursing
which may dictate taking additional STEM classes and having such role
models. Moreover, no respondents chose engineering (C) careers.
The Linear regression showed that the more STEM classes and role
models a female had, she tended to earn a higher salary which also
correlated with choosing a STEM career in science or math (A) or IT (B)
based on the ANOVA from questions one and two. However, the
relationship was not a strong one because some careers like nontechnical (D)
that include business and legal require a great deal of IT and math courses
and some caring professions (E) such as nursing and medical, also require a
great deal of math and science. The relationship for the number of STEM
classes and the impact on salaries was more significant than the number of
STEM role models and its influence on salaries.
The main issues with the hypotheses for research questions 2 and 3
was although the data showed a weak and direct relationship between the
variables, the relationship was insignificant because the sample size was too
small, creating larger residuals and lore standard error. However, STEM
classes and salaries was still more significant than role models and salaries.
The only one that showed significance was research question 1. For this
reason, the scatter plots in chapter 4 showed no significant relationship for
all three research questions.
Assumptions
I assumed in this study that there were females who were not always
encouraged to take more STEM classes and careers in school, or had not had
many female role models that could encourage her to take more STEM
classes and choose a STEM career (Correll, 2004; Eccles & Wang, 2016;
Wrigley, 2002). I also assumed that when contacting the alumni associations
of the four sampled school strata, that there would be reasonable cooperation
between the alumni association and myself in disseminating the surveys to
the students as randomly as possible. The associations were asked to contact
the students due to confidentiality, which they provided the link randomly to
alumnae. In this system I used for my study, each student had an equal
chance of participating (Shao, 2002). I did receive reasonable cooperation,
although there was one school that was more hesitant and I had to rely on
less data from that particular school. However, the other schools in the
sample offered more cooperation and had I had more cooperation, I would
have been able to survey a larger sample and perhaps this would have made
the relationship among the variables stronger. This reduced cooperation
resulted in a much smaller sample than the original one from chapter 3.
Moreover, the result was a small data set, resulting in larger residuals and
increased standard error.
Limitations
Some limitations that I had with my research included that the sample
was limited to only four Long Island universities, difficult to obtain a cross
section of the total population, based on a localized area, with a limited
geographic scope. The sample size was considerably smaller than I had
proposed in chapter 3. Originally I had forecasted an effect size for a sample
size in the 400s and I only received responses from 48 including the 4 from
the pilot respondents. For this reason, I have very small data sets. This made
it difficult to generalize to the entire population affecting external validity
when I conducted the analysis (Nachmias & Nachmias, 2008). Moreover,
the response rate was low and I only received a total of 48 responses, 4 for
the pilot and 44 for the general study. This resulted in a very small dataset
which made it difficult to generalize about the total population, resulting in
additional residuals and increased standard error. Furthermore, this was a
correlation study, which means that causation cannot be determined. I could
not claim that the lack of STEM classes that females took correlates with
them to choose careers outside of the STEM fields. I could only hypothesize
that there was a positive relationship with choosing STEM careers with the
number of STEM classes and female role models they had. I was also able
to hypothesize that there was a positive relationship between them taking
more STEM classes, and having more female role models with receiving
higher salaries. However, this is not always true as role models are only
remotely related to choosing STEM careers and salaries. For this reason, I
could only marginally accept the alternative hypothesis for all three
questions. I could not accept the null hypothesis because although the
differences were small and insignificant, they did exist. Moreover, I did not
obtain as much cooperation as I would have liked and therefore had a small
sample size. A broader geographic sample and a larger sample would
delineate this limitation.
Other limitations were my financial and mobility constraints. For
these reasons, it was necessary to conduct the study online using an online
survey instrument. I must make sure the questions are objective and as valid
as possible with a Cronbach alpha of.7 or .8 (Field, 2013). My Cronbach
alpha was lower than I might have needed to make some adjustments to the
questions to reduce bias.
The categories of the factors also presented limitations. Originally I
had seven groups but due to low response, I had to reduce them to 5. Also,
the manner to which I created the categories was limited because not only
STEM careers (category A or 1) take STEM classes and have STEM role
models. Also, nontechnical (category D or 4) careers like business and legal
or accounting also take a large number of math and IT (category B or 2)
classes. Moreover, even some caring professions like nursing or medical
(category E or 5) take a large number of STEM classes, especially in the
biological sciences. There was also difficulty in representing proper scatter
plots because of the small data sets. Using SPSS, I had to use numbers to
coordinate with the letters as seen above. There were no responses for
category C or 3 which was engineering.
Delimitations
As stated in chapter 1, delimitations are the factors that I as the
researcher have chosen which are the boundaries I have set for this study. In
restating my boundaries, the first boundary I have set is that I am only
considering females born after 1980, living in Long Island who was alumnae
of the four universities chosen for this study, making this a stratified random
sample, through an online survey. The reason this study was online was to
control cost and also because I have difficulty with mobility and require
personal assistance to mobilize. Moreover, online surveys are easier to
administer, more global, cost effective, and have higher response rates than
postal mail surveys (Patton, 2009; Shao, 2002).
Implications
As stated in Chapter 1, if the results of the study demonstrated that
taking more than three years of STEM classes in high school and
postsecondary school and having role models correlate positively with
career choices, this could help females obtain the training necessary to
impact their decisions to pursue these career choices. However, while the
results do demonstrate this relationship, only the number of STEM classes
and career choices is a strong one.
A major benefit for females could be higher pay as a result of being
able to make career choices in the STEM fields. This is a practical benefit
because females need to pay bills, earn a living, and save for retirement. If
females were given more opportunities to take STEM classes, then females
would able to increase their representation in STEM fields as postulated by
Carrell, Page, and West (2010), Farland-Smith (2009), Gilligan (1988),
Noddings (1986) and Sharp, et al, 2008). Moreover, by adhering to
Erikson’s theory, by encouraging young girls to be interested in STEM
through play and learning as children, perhaps more of these females will
choose STEM careers, later in life. I had hoped that this study will denounce
Acker’s theory by showing that it is not unnatural for females to enter
STEM professions that were a traditionally male dominated, but that they
just did not have the opportunities that their male counterparts had, which is
slowly changing. Moreover, females are just as naturally capable to succeed
in STEM as their male counterparts (Carrell, Page, & West, 2010), Farland-
Smith, 2009).
Recommendations for Future Action
Here is some further research using correlations to test further of what
was already being studied.
ANOVA comparing career choices and STEM classes.
For the comparison among the five categories of career choices and
the relationship between the numbers of STEM classes this sample of
females in this small data set took showed a positive, significant relationship
between the two variables. The relationship of STEM classes and career
choices was much stronger than the career choices and role models, hence
the p value of .000 in table 1 in chapter 4. However, when looking in more
detail from table 3 comparisons in chapter 4, despite the .000 p value, the
only category that appeared to be significant in the relationship between
career choices and the number of STEM classes was the nontechnical
category. This may be due to the fact that even business majors, and law
majors have to take science and math classes. There was also a significant
difference between the number of STEM classes and caring professions
most probably because some caring professions may include nursing or
healthcare which requires science and math. There was no significance
between the other categories especially between math and science and IT.
For the role models, there does not seem to be a significant difference
between the career choices and the number of female STEM role models
across career groups especially since this is a small data set therefore,
additional research is needed. For this reason further research is needed with
a larger data set.
Moreover, the resulting p value in this case was .000 possibly because
even some of the caring professions like nursing or medical take a large
number of STEM classes and have additional female role models as do
females in nontechnical professions like business or legal where math is
required, reducing the significance slightly (Field, 203). Therefore, there is a
significant difference between the number of STEM classes and the career
choices among the career choice groups, but mostly between categories A or
1 (science) and D or 4 nontechnical). This first ANOVA was a general one
using both STEM and Role which is shown in the Appendix. Then to answer
each of the first research questions and determine those hypotheses, I
conducted separate ANOVAs, one with career choices and STEM and the
other with career choices and role to determine a true significance. The first
ANOVA table showed the significance with career choices and both the
number of STEM courses and role models together. The second two show
the individual ANOVAs based on the first two research questions and
hypotheses.
ANOVA comparing career choices and role models.
For role models versus career choices, the table in chapter 4 showed
that the total sum of squares is 37.27, the degree of freedom intergroup is 3
and the mean squares across groups are 1.78. Moreover the significance
is .08 which is slightly above .05 making the difference between the number
of role models and one’s career choices, less significant. In other words,
there was not a significant relationship with the number of role models one
had in school and whether or not the respondent chose a career in math and
science, IT, nontechnical, or a caring profession. Therefore more research
needs to be conducted with a larger data set.
Objectives related to STEM classes and role models.
In viewing the multiple comparisons from the ANOVA table 3 in
chapter 4 among the career choice groups and the significance between each
career choice group and the number of STEM classes and female STEM role
models, the only category that appeared to be significant in the relationship
between career choices and the number of STEM classes was the
nontechnical category and to a lesser degree, math and science. It seemed
that there is a weak and positive relationship between the number of STEM
classes and the number of STEM female role models. In other words, the
more science classes one took specifically, there seemed to be a slightly
higher number of role models.
Moreover, there appeared also to be a slightly higher number of role
models for nontechnical careers who tended to take additional STEM
classes, may be due to the fact that even business majors, and law majors
have to take science and math classes. There was also a significant
difference between the number of STEM classes and caring professions
since some caring professions may include nursing or healthcare which
requires science and math. There was no significance between the other
categories especially between math and science and IT. For the role models,
there did not seem to be a significant difference between the career choices
and the number of female STEM role models across career groups.
Number of STEM classes differ base from number of role models
In general, the number of STEM classes was related to the career
choice and also dictated by the major of the student and in high school, by
the school district requirements. As for the female STEM role models, these
varied slightly based on the number of female STEM role models a female
has in her family, whether her science or math or IT teachers in high school
or postsecondary were females and influenced her or whether a female
doctor inspired her.
Correlation between number of STEM classes and salaries
The correlation between the number of STEM classes and salaries
tended to be a weak positive one, but stronger and more significant than
STEM role models and salaries. In other words, the more STEM classes,
especially science and math, that a female took in high school and
postsecondary education, generally the higher salary she earned. relationship
was not a strong one because some careers like nontechnical (D) that include
business and legal require a great deal of IT and math courses and some
caring professions (E) such as nursing and medical, also require a great deal
of math and science. Therefore the relationship among the variables was not
very significant, resulting in a marginal acceptance of the alternative
hypothesis, due to the small data set.
Correlation between number of STEM female role models and
salaries The correlation between the number of STEM female role
models and salaries was a weak positive correlation. Since the relationship
between the number of role models and career choices was a weak positive,
meaning that in general females that chose a STEM career, females who
tended to take more STEM classes, had slightly more role models while
earning somewhat higher salaries. However, the results were insignificant
between the number of role models and salaries as was the relationship
between the number of role models and career choices. Therefore, the
alternative hypothesis was marginally accepted.
Recommendation for Further Research
I recommend for action, that since the relationship between the
number STEM role models with career choices and salaries had weak
positive relationships with little significance for the most part, further study
must be conducted. The relationship between STEM classes and career
choices was most significant. The relationship of STEM classes and salaries
was more significant than role models, but with a small data set, additional
research with a larger data set would benefit the body of literature.
Moreover, as ByarsWinston, (2014) has indicated, the government needs to
continue to expand investment in helping females to enter STEM careers
through education, and career development. Perhaps in future research, a t
test can be conducted with career choice categories A or 1 science and math
and D or 4, nontechnical because nontechnical fields include business, legal,
and social sciences, many who take a considerable number of STEM classes
particularly math, science, and IT. Even some caring professions need a
considerable number of STEM classes if the career is nursing, physician
assistance, or healthcare.
Furthermore, as I mentioned in my chapter 5 limitations, I was not
able to obtain as much broad cooperation as would have been ideal from the
sample universities. Hence, my sample size was substantially reduced,
resulting in a small data set. Therefore, I recommend for further study, using
a larger sample of universities, resulting in a larger sample size of alumnae.
A larger sample and a broader geographic area may offer more valid and
reliable results with fewer residuals and deviations from the mean, which
will result in increased validity, reliability, and reduced residuals and
deviation.
