American Educational Research Journal

October 2013, Vol. 50, No. 5, pp. 1081–1121

DOI: 10.3102/0002831213488622
Ó 2013 AERA. http://aerj.aera.net

Why Students Choose STEM Majors:

Motivation, High School Learning, and
Postsecondary Context of Support
Xueli Wang
University of Wisconsin-Madison

This study draws upon social cognitive career theory and higher education
literature to test a conceptual framework for understanding the entrance
into science, technology, engineering, and mathematics (STEM) majors by
recent high school graduates attending 4-year institutions. Results suggest
that choosing a STEM major is directly influenced by intent to major in
STEM, high school math achievement, and initial postsecondary experiences,
such as academic interaction and financial aid receipt. Exerting the largest
impact on STEM entrance, intent to major in STEM is directly affected by
12th-grade math achievement, exposure to math and science courses, and
math self-efficacy beliefs—all three subject to the influence of early achieve-
ment in and attitudes toward math. Multiple-group structural equation
modeling analyses indicated heterogeneous effects of math achievement
and exposure to math and science across racial groups, with their positive
impact on STEM intent accruing most to White students and least to under-
represented minority students.

KEYWORDS: STEM participation, college major choice, social cognitive career

theory, multiple-group SEM

Introduction

W ithout question, America’s ability to maintain its global competitive-

ness within science, technology, engineering, and mathematics
(STEM) fields is an issue of national importance. Often framed in the context
of human capital (National Science Board, 2010), discussions of the critical
issues facing the nation’s STEM infrastructure center on a recognized need

XUELI WANG is an assistant professor in the Department of Educational Leadership and

Policy Analysis at the University of Wisconsin-Madison, 270-H Education Building,
1000 Bascom Mall, Madison, WI 53706-1326; e-mail: [email protected]. Her
research interests include participation in STEM fields of study and pathways and suc-
cess of students beginning at community colleges.
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
for building STEM workforce capacity (National Academies 2005 ‘‘Rising
Above the Gathering Storm’’ Committee, 2010). Support for this cause has
been levied through investments in educational programming, many of
which are focused on postsecondary education.
The demand for graduates in STEM fields continues to grow at a relatively
rapid rate. According to the National Science Foundation (2010), the employ-
ment rate in science and engineering fields rose an average of 3.3% annually
between 2004 and 2008 compared to an average 1.3% annual increase in
employment in all occupations, and this estimated growth rate is consistent
with long-term national trends (U.S. Department of Labor, 2007). By 2018, 9
of the 10 fastest growing occupations that require at least a bachelor’s degree
will depend on significant math or science training, and many science and
engineering occupations are predicted to grow faster than the average rate
for all occupations (Lacey & Wright, 2009; National Science Board, 2010).
These data document the need for greater participation of qualified col-
lege graduates in the STEM workforce. However, the supply side of the
STEM pipeline still reports a serious shortage of students pursuing STEM dis-
ciplines (Fox & Hackerman, 2003). While the national demand for motivated
students to enter postsecondary STEM fields is at its highest, high school
seniors’ interest in and readiness for pursuing these majors have been slug-
gish (ACT, 2006). American postsecondary institutions are therefore facing
an unprecedented need to increase the number of students who study in
STEM disciplines.
Of particular concern in the discussion on broadening STEM participation
is the underrepresentation of racial minorities, women, and students of low
socioeconomic status (SES; e.g., Anderson & Kim, 2006; Herrera & Hurtado,
2011; National Academies 2005 ‘‘Rising Above the Gathering Storm’’
Committee, 2010; National Science Foundation, 2006, 2010; Schultz et al.,
2011). An overwhelming body of research has also suggested that underrep-
resented racial minorities, women, and students of low SES persist at lower
rates in STEM fields of study than their White, male, and more socioeconom-
ically advantaged counterparts (e.g., Bailyn, 2003; Blickenstaff, 2005; Kulis &
Sicotte, 2002). It has been established that college majors create differential
opportunities for social mobility and that college graduates from STEM fields
attain higher occupational earnings and social status positions associated with
these professions compared to many other fields (Russell & Atwater, 2005). In
this sense, the differential participation rates in STEM fields are particularly
detrimental because they adversely affect those underrepresented students’
long-term social mobility, thus perpetuating socioeconomic inequality
(Carter, 2006). Therefore, the shortage of these students successfully pursuing
and completing studies in STEM disciplines continues to be a significant con-
cern for educators, policymakers, and researchers alike.
Although these rising calls have generated a fair amount of empirical
interest, most research concentrates on persistence and attainment among

1082
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
students who have already entered STEM fields. Not enough attention has
been paid to factors relevant to interest in and entrance into STEM fields,
which are arguably the first critical steps into the STEM pipeline. Given the
previously discussed pressing concerns facing STEM education nationally, it
is pivotal to provide rigorous academic programs and support mechanisms
that prepare students, especially members of traditionally underrepresented
groups, to enter these challenging and important fields of postsecondary
study. Needless to say, this educational endeavor will rely on collective, con-
certed, and well-informed efforts by the nation’s educational institutions. A
decision to pursue a STEM major is a longitudinal process that builds during
secondary education and carries into postsecondary studies. A full picture of
this process is best realized through incorporating the effects of these two lev-
els of education since they both shape students’ entrance into STEM. Treating
secondary and postsecondary education effects in isolation would severely
limit the ability to fully make sense of this phenomenon. As such, theoretically
based work from a holistic, K–16 perspective is needed to better understand
boosters and barriers to students’ entrance into STEM fields of study. Toward
that end, a theoretical model of STEM participation is proposed and tested in
this study focusing on both secondary and postsecondary factors. Particular
attention is also given to the potentially varying effects of these factors among
different student subpopulations by analyzing multiple-group structural equa-
tion models based on race, gender, and SES.

Background Literature and Theoretical Framework

Research on STEM Education
STEM education has garnered close scholarly attention. Numerous studies
have revealed the disproportionately high attrition rates of women and minor-
ities and the bachelor’s degree completion gap in STEM disciplines at 4-year
institutions across the nation (e.g., Anderson & Kim, 2006; Huang, Taddese,
& Walter, 2000; Seymour & Hewitt, 1997). In addition to the gender and racial
disparities in STEM persistence and completion, researchers also have high-
lighted theoretical reasons that students persist or leave a STEM field of study,
such as early exposure to and proficiency in math and science (Adelman,
1998, 1999, 2006; Anderson & Kim, 2006); high school curriculum (Elliott,
Strenta, Adair, Matier, & Scott, 1996); advanced courses in math and science
(Ellington, 2006); information early in the career search process (Holland,
1992); the types of opportunities, experiences, and support students receive
in college (e.g., M. J. Chang, Sharkness, Newman, & Hurtado, 2010;
Seymour & Hewitt, 1997); institutional selectivity (M. J. Chang, Cerna, Han,
& Sáenz, 2008; Eagan, 2009; Strayhorn, 2010); faculty quality and diversity
(Brainard, Metz, & Gillmore, 1993; Leach, 2010); and classroom experiences
(Cabrera, Colbeck, & Terenzini, 2001).

1083
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Despite the wealth of research on persistence and completion in STEM
fields, less focus has been given to entrance into postsecondary STEM disci-
plines. Existing research does reveal that the choice to pursue STEM fields is
affected by math- and science-related interest and self-assessment (e.g.,
Seymour & Hewitt, 1997), math and science completed during high school
(e.g., Ethington & Wolfle, 1988; Maple & Stage, 1991), social background
(Ware & Lee, 1988), and parental education (Gruca, Ethington, &
Pascarella, 1988). The most comprehensive national study to date on stu-
dents who enter STEM was conducted by Chen and Weko (2009).
Utilizing three Institute of Education Sciences (IES) longitudinal data sets,
the authors found that the percentage of students entering STEM fields
was higher among male students, younger students, students financially
dependent on family, Asian/Pacific Islander students, foreign students, or
those who spoke a language other than English as a child, and students
with more advantaged family background and stronger academic prepara-
tion than their counterparts. However, given the descriptive nature of the
study, factors influencing STEM entrance beyond demographics were barely
examined. Another recent study (Crisp, Nora, & Taggart, 2009) found that
students’ decisions to declare a STEM major and earn a STEM degree at
a Hispanic-serving institution were influenced by their gender, ethnicity,
SAT math score, and high school class rank percentile. Despite these com-
mendable empirical efforts, relatively less is known at the national level
about why students enter STEM fields.
Overall, research on STEM education represents substantial empirical ef-
forts to form a better understanding of the underlying factors that influence
student success along the STEM pipeline. Yet few academic studies using
nationally representative samples have dealt with the very first step of
STEM participation: why students enter STEM majors. The primary focus
of existing studies based on national samples revolves around students
who have already chosen a STEM major (e.g., M. J. Chang et al., 2008,
2010; Eagan, 2009). Furthermore, while abundant data exist to indicate the
low enrollment and high attrition rates in STEM fields of racial minorities,
women, and students of low SES, little is known in regard to how factors
influencing STEM entrance work differently or similarly across these sub-
groups of students.
Aside from the imperative need for adding to the empirical knowledge
base on STEM entrance, research in this vein also calls for a new theoretical
framework that holistically and longitudinally captures supports and barriers
to students choosing STEM majors. Indeed, as previously noted, a small
body of research has looked at the issue of STEM enrollment, yet these stud-
ies either are heavily focused on secondary school and background influen-
ces (Maple & Stage, 1991; Tyson, Lee, Borman, & Hanson, 2007) or solely
deal with the fit between postsecondary disciplinary environments and stu-
dents’ interests (Olitsky, 2012; Toker & Ackerman, 2012), often in isolation of

1084
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
each other. Although these studies are well grounded in prior literature, their
theoretical considerations provide limited insight illustrating one or only
a few aspects of the issue and do not explicitly account for the developmen-
tal and longitudinal nature of a student’s interest in and decision to pursue
a particular field of postsecondary study. In addition, important postsecond-
ary supports and barriers such as financial aid, academic interaction, and
remediation that could influence STEM entrance after students enroll in col-
lege are seldom addressed in those frameworks. Recognizing these research
gaps and the lack of a comprehensive framework on STEM entrance in the
literature, this study draws on a theoretical model with an intentional
emphasis on the secondary-postsecondary nexus of the STEM pathway
that accounts for the holistic and longitudinal nature of STEM entrance. A
detailed discussion of this framework follows.

Theoretical Framework
The theoretical model (Figure 1) integrates the social cognitive career
theory (SCCT) and prior literature on factors closely related to college stu-
dents’ academic choices and outcomes. In this model, students’ intent to
major in STEM is affected by their 12th-grade math achievement, exposure
to math and science courses, as well as math self-efficacy beliefs, all of which
are subject to the influence of prior achievement in and attitudes toward
math. Students’ STEM intent in turn affects their actual choice of STEM fields
of study. In addition, entrance into STEM fields also is directly influenced by
postsecondary context of supports and barriers. To be specific, postsecond-
ary supports include academic interaction, financial aid, college readiness in
math and science, graduate degree expectations, and enrollment intensity.
Among postsecondary barriers are remediation (taking remedial courses in
math, reading, and writing) and external demands such as having children
and the number of work hours. A more detailed description of the model’s
theoretical grounding and supporting literature follows.
Based on Bandura’s (1986) general social cognitive theory, SCCT under-
scores the interrelationship among individual, environmental, and behav-
ioral variables that are assumed to undergird one’s academic and career
choice (Lent & Brown, 2006). Key factors in SCCT include self-efficacy be-
liefs, outcome expectations, interests, environmental supports and barriers,
as well as choice actions (Lent, Sheu, Gloster, & Wilkins, 2010). SCCT offers
an appropriate theoretical lens to study the issue of STEM choice (Lent,
Brown, & Hackett, 1994, 2000) and has been applied in a small number of
studies on STEM-related academic choice intentions (e.g., Betz & Hackett,
1983; Byars-Winston, Estrada, Howard, Davis, & Zalapa, 2010; Hackett,
Betz, Casas, & Rocha-Singh, 1992; Lent, Lopez, & Bieschke, 1993; Lent,
Lopez, Lopez, & Sheu, 2008). Although this set of studies suggests the valid-
ity of SCCT as an explanatory framework for understanding STEM interests

1085
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang

Postsecondary Context of Supports

College
Academic Readiness in
Interaction Math and
Science
High School
Senior Year Expecting a
Financial Enrollment
Early High Graduate
Math Aid Intensity
School Degree
Self-Efficacy
Beliefs

Attitudes
Toward Math Exposure to Intent to Entrance into a
10th Grade Math and Major in a STEM
Science STEM Field Field of Study

Math
Achievement
10th Grade Math Postsecondary Context of Barriers
Achievement
Remediation External Demands
12th Grade

Children
Reading

Having
Writing

Hours
Work
Math

Secondary Postsecondary

Figure 1. Theoretical model for the study.

and choices, they are largely limited by cross-sectional designs and single-
institution data (Lent et al., 2010). Based on a national longitudinal database,
this study incorporates the key constructs of SCCT to build a conceptual
model of STEM participation and capture the nature of the relationships
among the theoretical variables over time.
SCCT posits that determination to produce a particular choice can be ex-
plained as a result of interests and goals. Therefore, choosing a STEM major
is hypothesized to be influenced by students’ intent to pursue these fields
upon postsecondary entry. Meanwhile, based on SCCT, interest in a choice
action is subject to self-reference belief and learning experiences. Given the
fundamental importance of early math experience in future STEM education
(e.g., Adelman, 1999; Bowman, 1998; Marshall, McGee, McLaren, & Veal,
2011; National Science Board, 2004), STEM intent can thus be argued as
a product of motivational attributes and learning as related to math at the
secondary level. More specifically, this intent is related to high school
seniors’ math achievement, exposure to math and science courses, and

1086
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
math self-efficacy beliefs (i.e., individuals’ confidence in their ability to suc-
cessfully perform or accomplish math tasks or problems; Hackett & Betz,
1989; Pajares & Kranzler, 1995). Furthermore, these three elements are
shaped by early math achievement and attitudes, especially in light of the
longitudinal and developmental nature of achievement in and attitudes
toward math (Eccles, 1994; Trusty, 2002).
SCCT also highlights the role of environmental supports and barriers in
determining choice actions. In a postsecondary setting, students’ pursuit of
STEM as an academic goal responds to contextual supports and barriers—
social, academic, or financial. Students transitioning into postsecondary edu-
cation navigate a series of demands, such as the need for financial resources,
academic integration into college, and various external demands. The out-
comes of this process might present either supports or barriers and thus
impact students’ academic choice behavior. Therefore, the proposed con-
ceptual model also includes a number of supports and barriers in this tran-
sition process, discussed in the following paragraphs.
Postsecondary supports are represented by academic interaction, col-
lege readiness in math and science, financial aid receipt, expecting a gradu-
ate degree, and enrollment intensity. Academic interaction between students
and other college socialization sources, such as faculty and academic advi-
sors, positively influences numerous student outcomes (Astin, 1993; J. C.
Chang, 2005; Terenzini, Pascarella, & Blimling, 1999). Such interactions
may provide necessary support for students to clarify and confirm their
choice of major field of study. Also, as K–12 assessments are not always in
perfect alignment with the academic requirements of postsecondary institu-
tions (Goldrick-Rab, Carter, & Winkle-Wagner, 2007), once in college, stu-
dents’ perceptions of the extent to which their high school math and
science courses have prepared them for college-level work may influence
their decision to pursue STEM. Students who feel that they are college-ready
in the areas of math and science may favorably consider a STEM major. In
addition, the receipt of financial aid affects students’ academic choices
(e.g., DesJardins, Ahlburg, & McCall, 2006; Ishitani & DesJardins, 2002)
and in particular may positively influence students’ choice of a STEM major
(Kienzl & Trent, 2009).
The conceptual model also includes enrollment intensity and graduate
degree expectations. Enrollment intensity—whether students enroll full-
time or less than full-time—often indicates the amount of time and psycho-
logical energy students devote to their educational experience (Wang, 2009)
and is positively linked to a number of postsecondary outcomes (Berkner,
Cuccaro-Alamin, & McCormick, 1996). Also, degree aspirations are strongly
related to educational choices and outcomes (Carter, 2002; Pascarella &
Terenzini, 2005; Wang, 2013). Although not necessarily providing direct, tan-
gible structural support to STEM entrance, these two elements may indicate

1087
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
the level of students’ psychological commitment to their studies and should
be accounted for in understanding student entrance into STEM majors.
In regard to postsecondary barriers to STEM entrance, the proposed the-
oretical model includes remediation and external demands. For many stu-
dents, remediation is a necessary part of the curriculum (Pascarella &
Terenzini, 2005). However, research on the effect of enrolling in remedial
courses has produced mixed results (Adelman, 1999; Bahr, 2008; Bailey &
Alfonso, 2005; Long, 2005). In examining the relationship between remedi-
ation and student choice of STEM, this study will provide targeted, context-
based research evidence regarding the effectiveness of remediation in sus-
taining students’ academic aspirations. In addition, the external demands
that students may need to deal with, for example, having dependent chil-
dren and working long hours, may redirect them from pursuing challenging
fields of study such as STEM. Together, these initial college experiences at
students’ first postsecondary institution are presumed to directly shape their
decisions to pursue STEM fields of study.
As previously argued, sociodemographic differences are of critical
importance in STEM-related research (Crisp et al., 2009), and persistent gen-
der and racial gaps in the STEM pipeline remain (Anderson & Kim, 2006;
Clewell & Campbell, 2002; Dowd, Malcom, & Bensimon, 2009). This war-
rants the need for STEM-related research to take such background differen-
ces into consideration. In this study, the proposed theoretical framework is
assessed separately across racial, gender, and SES groups (more details pro-
vided in the methods and results sections of the article). This approach not
only helps evaluate the framework’s applicability across student subpopula-
tions, but also illuminates how the proposed relationships in the model may
differ based on race, gender, and SES.

Research Questions
Guided by the conceptual framework, this study examines the direct
and indirect influences of high school exposure to math and science,
achievement and motivational attributes as related to math, and initial post-
secondary experiences on entrance into STEM fields of study in college.
Specifically, this research addresses the following interlocking questions:

Research Question 1: What are the relationships among high school exposure to
math and science, achievement and motivational attributes as related to math,
intent to pursue STEM upon entry into postsecondary education, and entrance
into STEM fields of study?
Research Question 2: Taking into account the relationships described in Question
1, how are students’ initial postsecondary education experiences, such as aca-
demic interaction, receipt of financial aid, and remediation, related to STEM
entrance?
Research Question 3: How do these relationships vary by race, gender, and SES?

1088
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
Methods
Data Source and Sample
Data for this study came from the Education Longitudinal Study of 2002
(ELS:2002), which was designed to study the transition of young people from
high school into postsecondary education and the workplace. ELS:2002
started with a nationally representative cohort of high school sophomores.
The sample was then augmented in the first follow-up study in 2004 to rep-
resent high school seniors. In 2006, roughly 2 years after high school, the
second follow-up study collected data on access to postsecondary institu-
tions, choices of enrollment and college major, and other aspects of college
experience. Given its focus on the transition from high school to postsec-
ondary education, ELS:2002 was an appropriate data set for this study. To
fully understand student learning, motivation, interest, and choice as related
to STEM majors, it is necessary to follow the same individuals from second-
ary to postsecondary education. The longitudinal data from ELS:2002 pro-
vided a thorough empirical description of student experiences relevant to
STEM education in high school and early years of college. (For complete
information on ELS:2002, see http://nces.ed.gov/surveys/els2002/.)
This study focused on the spring 2004 high school graduates who had
enrolled in a postsecondary institution by 2006. Of approximately 14,000
members of the 2004 senior cohort, about 12,500 (89.3%) responded to
the second follow-up interview. For the purpose of this study, an initial total
of 6,300 (out of 12,500 eligible) students who reported postsecondary atten-
dance at a 4-year institution by 2006 were retained. Among these students,
roughly 19.3% intended to major in STEM upon entering college while
80.7% were interested in other fields of study; 15.4% (out of all 6,300
4-year enrollees) declared a major in a STEM field by 2006, compared to
84.6% who chose other disciplines or had not declared a major. All analyses
were weighted using the appropriate ELS panel weight (F2F1WT).

Measures
This section summarizes variables that were included in the study based
on the theoretical model. The main outcome, entrance into STEM, was
a dichotomously coded variable based on the survey item that asked re-
spondents’ field of study during the 2006 ELS second follow-up interview.
The focal mediating variable was intent to pursue a STEM field, measured
by whether the most likely postsecondary field of study students considered
upon postsecondary entry was in a STEM discipline.
Five variables at the secondary school level were included:1 (a) expo-
sure to math and science courses, measured by the number of units in math-
ematics and science technologies that students took; (b) 12th-grade math
achievement, measured by math standardized test scores at the 12th grade;2

1089
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
(c) 12th-grade math self-efficacy beliefs, measured by five items—each on
a 4-point Likert scale—that represented students’ self-efficacy beliefs in areas
such as taking math tests, mastering math skills, and completing math assign-
ments; (d) 10th-grade math achievement, measured by math standardized
test scores at the 10th grade (see Note 2); and (e) 10th-grade attitudes toward
math, measured by three items—each on a 4-point Likert scale—that repre-
sented students’ perceived enjoyment and importance of math.
Although these variables measured during high school offered some
insight into student learning in math and science, they did not indicate fully
how well such learning prepared students for college-level work. To provide
a more comprehensive picture that went beyond course-taking and achieve-
ment, a latent variable at the college level was included that measured col-
lege readiness in math and science: the extent to which college students
believed that their high school math and science courses prepared them
for college-level work.
Also included to represent postsecondary context of supports and barriers
were: academic interaction, receipt of financial aid, enrollment intensity, gradu-
ate degree expectations, remediation, and external demands. Academic interac-
tion was measured by the frequency of interacting with faculty about academic
matters, meeting with advisors about academic plans, and working on course-
work at school libraries. Receipt of financial aid was a dichotomous variable
based on students’ first-year aid status. Enrollment intensity was measured by
a dichotomous variable indicating whether students’ college enrollment was
full-time or not. Similarly, the variable measuring graduate degree expectations
was dichotomous: coded 1 if students expected to earn a graduate degree and
0 otherwise. Remediation included three dichotomous variables: whether stu-
dents took remedial courses to improve reading, writing, and math skills.
Representing external demands were (a) one dichotomous variable measuring
whether students had dependent children and (b) a continuous variable measur-
ing the average number of weekly hours students worked for pay. Table 1 lists
the names, descriptions, and ELS labels of all variables used in the study. In the
table, each latent construct and its corresponding indicators also are specified.

Analysis
Descriptive Analysis
First, descriptive statistics were computed and disaggregated by the sam-
ple’s background characteristics. These descriptive statistics provided a gen-
eral profile of the ELS 2004 high school senior cohort’s participation in STEM
fields of study 2 years after high school graduation, and helped identify any
variation in STEM intent and entrance across sociodemographic variables.

1090
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Table 1
List of Variables in the Study

Variable Name Description Education Longitudinal Study Label

Endogenous variable
Entrance into STEM fields of Respondent’s 2006 major field of study is in STEM fields; 1 = yes, Recoded from F2MAJOR2
study (STEM) 0 = no
F2MAJOR4
Mediating variable
Intent to major in a STEM Field of study respondent would most likely pursue when Recoded from F2B15
field (INTENT) beginning at the first postsecondary institution is in STEM fields;
1 = yes, 0 = no
12th-grade math achievement High school senior math standardized score F1TXMSTD
12th-grade math self-efficacy Can do excellent job on math tests F1S18A
beliefsa Can understand difficult math texts F1S18B
Can understand difficult math class F1S18C
Can do excellent job on math assignments F1S18D
Can master math class skills F1S18E

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

Items based on 4-point Likert scales with 4 indicating almost
always and 1 indicating almost never
High school exposure to Units in mathematics from high school transcript F1RMAT_C
math and science coursesa Units in science from high school transcript F1RSCI_C

(continued)

1091
 Table 1 (continued)

1092
Variable Name Description Education Longitudinal Study Label

Exogenous variable
10th-grade math attitudesa Gets totally absorbed in math BYS87A
Thinks math is fun BYS87C
Mathematics is important BYS87F
Items based on 4-point Likert scales with 4 indicating strongly
agree and 1 indicating strongly disagree
10th-grade math achievement High school sophomore math standardized score BYTXMSTD

Postsecondary context of supports and barriers

Academic interactiona Talk with faculty about academic matters outside of class F2B18A
Meet with advisor about academic plans F2B18B
Do coursework at library F2B18C
Items based on 3-point scales with 3 indicating often and 1
indicating never
College readiness in math High school math prepared for college F2B17A
and sciencea High school science prepared for college F2B17B
Items based on 3-point scales with 3 indicating a great deal and 1
indicating not at all
Financial aid receipt Offered financial aid first year at college; 1 = yes, 0 = no F2PS1AID
Enrollment intensity 1 = full-time, 0 = less than full-time Recoded from F2PSIFTP

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

(FULLTIME)
Expecting to earn a graduate Whether respondent expected to earn a graduate degree; 1 = yes, Recoded from F2STEXP
degree (GRADEXP) 0 = no
Remediation Took remedial reading; 1 = yes, 0 = no F2B16A

(continued)
 Table 1 (continued)

Variable Name Description Education Longitudinal Study Label

Took remedial writing; 1 = yes, 0 = no F2B16B

Took remedial math; 1 = yes, 0 = no F2B16C
External demands Whether has biological children; 1 = yes, 0 = no F2D03
Hours worked weekly (WRKHRS) F2C26R
F2C31R
Demographic variables
Race Underrepresented minorities, Asian American, and White Recoded from F1RACE
Socioeconomic status (SES) The SES variable is quartile coding of the composite variable F1SES1QR
measuring SES in ELS:2002. This composite variable was
constructed based on five equally weighted, standardized
components: father’s/guardian’s education, mother’s/guardian’s
education, family income, father’s/guardian’s occupation, and
mother’s/guardian’s occupation (Source: ELS:2002 Electronic
Codebook).
Gender Respondent’s gender (male or female) F1SEX
a
A latent variable.

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

1093
Wang
Confirmatory Factor Analysis
Following the descriptive analysis, a two-step modeling approach was
adopted. First, the measurement part of the model was examined. If the
measurement model was acceptable, then the full structural equation model
was analyzed where the measurement and structural parts of the model were
simultaneously estimated (Kline, 2011).
A confirmatory factor analysis (CFA) was performed to analyze the pro-
posed measurement model that explicitly specified the latent factors and
their corresponding indicator items (Figure S1 in the online version of the
journal). Five latent constructs were measured: 10th-grade attitudes toward
math, 12th-grade math self-efficacy, high school exposure to math and sci-
ence, college readiness in math and science, and academic interaction. At
this step, fit statistics of the measurement model were assessed, and conver-
gent validity was checked by examining the standardized factor loa-
dings—the estimated correlations between individual factors and their
corresponding indicators (Kline, 2011). In addition, multiple-group CFA
were performed to evaluate whether the measurement model held across
subsamples.

Structural Equation Modeling

Following CFA, the proposed conceptual model was tested using struc-
tural equation modeling (SEM). Figure 2 is a depiction of the structural part
of the SEM diagram based on the theoretical model.
In mathematical form, the path structure in this study was postulated by
five simultaneously estimated regression equations. The first three equations
examined how 12th-grade math self-efficacy, exposure to math and science,
and 12th-grade math achievement were each influenced by 10th-grade math
achievement and attitudes toward math. The fourth equation investigated
how students’ intent to major in STEM was affected by 12th-grade math
self-efficacy, exposure to math and science, and 12th-grade math achieve-
ment. The final regression equation examined how students’ decisions to
enter into STEM fields of postsecondary study were affected by their intent
to major in STEM fields, college readiness in math and science, and postsec-
ondary context of supports and barriers (e.g., academic interaction, receipt
of financial aid, expectation of graduate degree, enrollment intensity, reme-
diation experience, and external demands). Moreover, students’ 12th-grade
math achievement was included in the final equation because math ability
might have a direct effect on students’ eventual entrance into STEM fields.3
In addition, given the strong relationship between math achievement and
math self-efficacy (Pajares, 1996; Pajares & Kranzler, 1995; Pajares & Miller,
1994), the SEM model specified that these two 12th-grade variables were cor-
related and this relationship was accounted for in the SEM analysis by esti-
mating the residual covariance between these two variables.4

1094
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors

Postsecondary Context of Supports

Expecting a
Financial Enrollment
Graduate
Aid Intensity
Math Degree
Self-Efficacy
Beliefs College
Attitudes 12th Grade Academic Readiness in
Toward
1 D
Interaction Math and
Math Science
10th Grade

High School Intent Entrance into a

Exposure to to Major in a STEM
Math and STEM Field Field of Study
Science
D
1
Math
Achievement 1 D
10th Grade
Remediation External Demands
Math
Achievement

Children
Reading

Writing

Having

Hours
Work
Math
12th Grade

Postsecondary Context of Barriers

2002 2004 2006

Figure 2. Diagram of proposed structural model for the structural equation mod-
eling analysis.
Note. To conserve space, the measurement part of latent factors (depicted as circles in Figure
2) is omitted from the path structural diagram. Exogenous variables are shaded; others are
endogenous variables. Note that endogenous variables, 12th-grade math self-efficacy, expo-
sure to math and science, math achievement, and STEM intent, serve as both a dependent
and an independent variable. D = disturbance term of the corresponding endogenous
variable.

The analyses were conducted using Mplus 6.1, a statistical software

package capable of SEM analysis that uses a mixture of different types of var-
iables (Kaplan, 2009; Kupek, 2006; Muthén & Muthén, 1998–2010). In addi-
tion, Mplus contains statistical tools that accommodate complex survey
design features such as survey weights and the clustering nature of
ELS:2002. Given the binary nature of the outcome variable in the fourth
and fifth equations previously described, probit regression models were
conducted for those two regression equations using the weighted least
square with adjustment in mean and variance (WLSMV) estimator. When
using the WLSMV estimator, the difference in chi-square values is not distrib-
uted as chi-square, so the DIFFTEST option in Mplus was used to obtain
a correct chi-square difference test between the baseline and nested models
(Muthén & Muthén, 1998–2010, p. 553).

1095
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Following Byrne (1998), the following fit indices were used to assess over-
all model fit: chi-square (x2), Comparative Fit Index (CFI), Tucker-Lewis Fit
Index (TLI), and root mean square error of approximation (RMSEA).

Multiple-Group Analysis: Testing for Structural Invariance

Following the full-sample SEM analysis, multiple-group analyses were
employed to examine whether the hypothesized model was equivalent
across subgroups. Specifically, this part of the study drew upon three sets
of analyses respectively based on race (Whites, Asians, and underrepre-
sented minorities5), gender (females and males), and SES (quartiles) and
tested for structural weight invariance across subgroups within each of these
three sociodemographic categories. Because this study focused on the struc-
tural pattern of the model (i.e., the underlying mechanism affecting students’
entrance into STEM fields of study), the model invariance tests concentrated
on the equivalence of structural path parameters across different groups.
To illustrate, in the gender-based multiple-group analysis, a baseline model
was first fitted—a multiple-group model with only factorial equality constraints
across gender, where the structural weights (i.e., regression coefficients) were
freely estimated across the male and female groups. Then, another multiple-
group model was estimated with cross-group constraints where all structural
weights across males and females were constrained to be equal. Next, a struc-
tural invariance test was conducted based on the corrected chi-square differ-
ence (Dx2) test that compared the baseline model with the constrained-equal
model. Through this test, if the corrected chi-square difference statistic did
not reveal a significant difference between the models, then it would be con-
cluded that the model had structural weight invariance across gender groups.
However, if non-invariance was indicated by a significant Dx2 statistic, then
structural weights (i.e., regression coefficients) were gradually constrained to
be equal across gender groups to determine whether group differences could
be attributable to any of the structural weights. If any constrained parameters
(i.e., structural weights) were found to be gender-invariant as suggested by
insignificant Dx2 statistics, then they would be constrained, cumulatively, in sub-
sequently more restrictive models. On the contrary, a significant Dx2 statistic
would suggest that the given parameter was not equivalent across gender
groups; therefore, it would be freely estimated in the subsequent models for
invariance tests (Byrne, 2010; Kline, 2011). Race- and SES-based multiple-group
analyses were carried out in the same fashion.

Indirect Effects
In this study, intent to major in STEM served as a mediator variable that
transmitted the effects of variables at the secondary level onto entrance into
STEM majors. In addition, 12th-grade math self-efficacy, exposure to math
and science, and math achievement were hypothesized to mediate the

1096
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
influence of 10th-grade math achievement and attitudes toward math on
intent to major in STEM. These indirect paths from high school variables
to STEM intent and to STEM entrance were estimated and the associated
indirect effects were calculated and tested for statistical significance using
Mplus’s MODEL INDIRECT command.

Missing Data
As is common with survey research, some of the variables included in
the study had missing data. In this study, Mplus’s full information maximum
likelihood (FIML) estimation was applied to handle missing data for the var-
iables that were treated as dependent by the software. Listwise deletion was
used to deal with the missingness in the exogenous observed variables.
Before performing listwise deletion, the data set that contained cases to be
deleted was compared with the data set that included cases not subject to
listwise deletion. It was observed that the distributions of variables in both
were quite similar. As a result, about 660 cases were removed from the anal-
ysis, resulting in the final analytic sample size of about 5,650.

Results
Descriptive statistics are presented in Table 2 to provide a comprehen-
sive picture of entrance into STEM majors based on student background
characteristics. The sample’s correlation matrices and means and standard
deviations for each measure are provided in Tables S2-S5 in the online ver-
sion of the journal. A discussion of the CFA and SEM model fit and the results
from multiple-group analyses follows. This section concludes with a descrip-
tion of the substantive results in light of the three research questions.

Results of Confirmatory Factor Analyses

The CFA analyses based on the whole sample as well as on racial, gen-
der, and SES subsamples indicated that the measurement model fit the data
well.6 RMSEA values in all models were well below the .05 cut-off point and
their 90% confidence interval upper bounds were all below .08, indicating
a good fit (Hooper, Coughlan, & Mullen, 2008; MacCallum, Browne, &
Sugawara, 1996). CFI and TLI values were all above .95, again suggesting
a good fit (Schreiber, Stage, King, Nora, & Barlow, 2006). All of the standard-
ized factor loadings were above .4 and significant at p \ .001, suggesting
good convergent validity of the measurement model (Kline, 2011). These
fit indices and factor loadings are presented in Table S1 in the online journal.

Results of Multiple-Group SEM Analyses

Prior to multiple-group analyses, the SEM model was analyzed based on
the whole sample, and fit indices suggested excellent model-to-data fit (line

1097
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 1098
Table 2
Descriptive Statistics of Demographic Characteristics of the Sample, Unweighted and Weighted

STEM Intent (2004) STEM Entrance (2006)

Total N Intended to Major in STEM Did Not Intend Declared a STEM Major Did Not Declare a STEM Major

N Wtd N N (%) Wtd N (%) N (%) Wtd N (%) N (%) Wtd N (%) N (%) Wtd N (%)

Total 6,300 1,560,050 1,220 (19.3) 302,860 (19.4) 5,090 (80.7) 1,257,180 (80.6) 970 (15.4) 240,670 (15.4) 5,330 (84.6) 1,319,370 (84.6)
Gender
Female 3,440 851,200 370 (10.9) 92,630 (10.9) 3,060 (89.1) 758,580 (89.1) 350 (10.2) 87,320 (10.3) 3,080 (89.8) 763,880 (89.7)
Male 2,870 708,840 840 (29.3) 210,240 (29.7) 2,030 (70.7) 498,610 (70.3) 620 (21.6) 153,360 (21.6) 2,250 (78.5) 555,490 (78.4)
Race/ethnicity
White 4,050 996,660 720 (17.7) 183,290 (18.4) 3,340 (82.3) 813,370 (81.6) 560 (13.9) 143,360 (14.4) 3,490 (86.1) 853,300 (85.6)
Asian 750 181,240 200 (26.7) 44,300 (24.4) 550 (73.3) 136,940 (75.6) 180 (24.0) 42,220 (23.3) 570 (76.0) 139,020 (76.7)
Underrepresented minorities
Black 660 166,090 140 (20.8) 35,170 (21.2) 530 (79.2) 130,920 (78.8) 110 (17.2) 30,840 (18.6) 550 (82.8) 135,250 (81.4)
Hispanic 520 135,480 100 (18.7) 25,860 (19.1) 430 (81.3) 109,620 (80.9) 70 (12.8) 14,470 (10.7) 460 (87.2) 121,000 (89.3)
American Indian 30 6,920 10 (25.0) 1,390 (20.0) 20 (75.0) 5,540 (80.0) 0 (7.1) 240 (3.5) 30 (92.9) 6,680 (96.5)
Multiracial 280 73,660 50 (18.5) 12,870 (17.5) 230 (81.5) 60,800 (82.5) 40 (14.9) 9,540 (13.0) 240 (85.1) 64,120 (87.0)
Socioeconomic status
Lowest quartile 700 173,830 160 (22.2) 37,250 (21.4) 550 (77.8) 136,580 (78.6) 90 (12.4) 20,150 (11.6) 620 (87.6) 153,680 (88.4)
Second quartile 1,050 268,580 180 (17.4) 47,940 (17.9) 870 (82.6) 220,640 (82.1) 140 (13.0) 34,610 (12.9) 910 (78.0) 233,970 (87.1)
Third quartile 1,620 398,640 270 (16.8) 69,080 (17.3) 1,350 (83.2) 329,560 (82.7) 230 (14.2) 56,240 (14.1) 1,390 (85.8) 342,400 (85.9)
Highest quartile 2,930 719,000 600 (20.6) 148,590 (20.7) 2,320 (79.4) 570,410 (79.3) 520 (17.6) 129,670 (18.0) 2,410 (82.4) 589,330 (82.0)

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

Note. The analytical Ns are rounded to the nearest 10 according to Institute of Education Sciences (IES) guidelines. The sum of subgroups may
not equal to the total in the table due to rounding. Wtd = weighted.
 Why Students Choose STEM Majors
1 of Table 3). Multiple-group SEM analyses were then conducted for racial,
gender, and SES groupings. Three sets of statistics and model fit indices were
derived from this series of analyses and are presented in Table 3. For exam-
ple, in the race-based multiple-group analysis, the hypothesized model was
initially fitted to the White, Asian, and underrepresented minority (URM)
samples separately. The fit statistics for the White-only model were
x2(275) = 1,063.85, relative x2 = 3.87, CFI = 0.980, TLI = 0.977, RMSEA =
0.028 (line 2 of Table 3); Asian fit statistics were x2(275) = 417.67, relative
x2 = 1.52, CFI = 0.974, TLI = 0.970, RMSEA = 0.028 (line 3); and URM fit sta-
tistics were x2(275) = 528.44, relative x2 = 1.92, CFI = 0.978, TLI = 0.975,
RMSEA = 0.026 (line 4). The fit indices suggested that the model fit each
racial group well. Thus, all racial groups were combined together and simul-
taneously fitted to the data to become the multiple-group baseline model.
The racial multiple-group baseline model also fit the data well: x2(881) =
1,987.54, relative x2 = 2.26, CFI = 0.979, TLI = 0.978, RMSEA = 0.026 (line 5).
In the next step, all of the 21 structural weights (i.e., regression coefficients)
were constrained equally across racial groups to examine structural weight
invariance across racial groups. The result of the corrected chi-square differ-
ence test was statistically significant (p \ .05), which suggested that one or
more of the parameters were non-invariant across racial groups (line 6).
Thus, instead of constraining these 21 parameters all at once, parameters
were constrained one by one to identify the source of non-invariance found
in the previous step. When the regression coefficient for the path from 10th-
grade math achievement to 12th-grade math self-efficacy was constrained
equal across racial groups, the result of the corrected chi-square difference
test was statistically significant, meaning that the regression coefficient of
this path was one of the sources of structural non-invariance across racial
groups (line 7). Similarly, non-invariance was found when the regression
coefficient for the following paths was constrained equally across racial
groups: from exposure to math and science to intent to major in STEM
(line 8) and from 12th-grade math achievement to intent to major in
STEM (line 9).
Furthermore, steps were taken to identify precisely where the non-
invariance of these three structural weights existed between specific pairings
of racial groups by performing partial constraints (i.e., selecting only two of
the three racial groups to be constrained equal at a time). Non-invariance
was found for all three pairs of comparison: White and Asian, Asian and
URM, and URM and White. Therefore, all three structural weights were freely
estimated across all racial groups in the model. Subsequent invariance tests
showed that there was no structural non-invariance caused by the remaining
18 structural weights.
The middle and lower sections of Table 3 display the model fit statistics
and multiple-group structural invariance test results for gender and SES
groupings, respectively. No structural non-invariance was found in the

1099
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Table 3
Racial, Gender, and Socioeconomic Status (SES) Multiple-Group Structural Weight Invariance Tests

1100
Tucker- Root Mean
Relative Corrected Comparative Lewis Square Error of
2
Model Description x (df) x2 (x2/df) Dx2(Ddf) Fit Index Fit Index Approximation

1 Single-group structural equation modeling (n = 5,650) 1,383.23 (275) 5.03 — .982 .979 .027
Racial groups
2 White 1,063.85 (275) 3.87 — .980 .977 .028
3 Asian 417.67 (275) 1.52 — .974 .970 .028
4 Underrepresented minorities 528.44 (275) 1.92 — .978 .975 .026
5 Baseline (factorial constrained) 1,987.54 (881) 2.26 — .979 .978 .026
6 All structural weights constrained 21 coefficients 2,002.84 (923) 2.17 75.05 (42)** .980 .979 .025
7 Constrained 10th-grade math achievement
! 12th-grade math self-efficacy 1 coefficient 1,990.68 (883) 2.25 8.28 (2)* .979 .978 .026
8 Constrained Exposure to math and science
! Intent to major in STEM 1 coefficient 2,002.60 (883) 2.27 4.00 (2)*** .979 .978 .026
9 Constrained 12th-grade math achievement
! Intent to major in STEM 1 coefficient 1,991.43 (883) 2.26 7.57 (2)* .979 .978 .026
10 All other 18 weights constrained (FINAL MODEL) 18 coefficients 1,979.15(917) 2.16 35.39 (36) .980 .980 .025
Gender groups
11 Male 795.27 (275) 2.89 — .981 .978 .027
12 Female 863.54 (275) 3.14 — .981 .978 .026
13 Baseline (factorial constrained) 1,681.42 (578) 2.91 — .981 .979 .026
14 All structural weights constrained 21 coefficients 1,671.16 (599) 2.79 28.79 (21) .982 .981 .025
SES groups
15 SES first quartile 392.94 (275) 1.43 — .974 .970 .026
16 SES second quartile 472.17 (275) 1.72 — .978 .975 .028

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

17 SES third quartile 573.71 (275) 2.09 — .981 .978 .027
18 SES fourth quartile 781.25 (275) 2.84 — .983 .980 .026
19 Baseline (factorial constrained) 2,291.35 (1,184) 1.94 — .981 .980 .026
20 All structural weights constrained 21 coefficients 2,308.68 (1,247) 1.85 71.40 (63) .982 .981 .025

Note. A significant Dx2 value indicates that the estimate is non-invariant across groups.
*p \ .05. **p \ .01. ***p \ .001.
 Why Students Choose STEM Majors
multiple-group analyses based on gender and SES, which indicated that the
hypothesized model can be operated equally across different subgroups
within gender or SES.

Final SEM Model

Through these detailed analyses, it became clear that a multiple-group
model based on race, where the paths from 10th-grade math achievement
! 12th-grade math self-efficacy, exposure to math and science courses
! intent to major in STEM, and 12th-grade math achievement ! intent
to major in STEM were freely estimated for all racial groups while all other
structural weights were constrained equal, was the most reasonable and via-
ble model. This final model fit the data, x2 (917) = 1,979.15, relative x2 = 2.16,
CFI = 0.980, TLI = 0.980, RMSEA = 0.025, and was slightly better in fit than
the baseline model. As for the parameter estimates, Table 4 displays the
direct and indirect effect estimates (both unstandardized and standardized)
from this final multiple-group model based on race. Changes in predicted
probabilities (CP) are also reported for significant coefficients in equations
modeling STEM intent and STEM entrance.
Figure 3 presents the final model with statistically significant paths high-
lighted, and the coefficient estimates are also denoted along with the paths.
The following summarizes specific results from the final model by
addressing the questions pursued in this study.

Research Question 1: What are the relationships among high school exposure to
math and science, achievement and motivational attributes as related to math,
intent to pursue STEM upon entry into postsecondary education, and entrance
into STEM fields of study?

Intent to pursue STEM was significantly and positively influenced by

12th-grade math self-efficacy, the effect of which remained the same among
all subgroups. The effect of high school exposure to math and science on
STEM intent was statistically significant and positive across all racial groups,
but was the smallest among underrepresented minority students. Math
achievement at the 12th grade was positively associated with intent to pur-
sue STEM fields among White students and underrepresented minorities, but
was null for Asian students.
Through intent to major in STEM, all three 12th-grade variables also indi-
rectly and positively affected actual choice of STEM majors after college
entry, except that 12th-grade math achievement did not show any significant
indirect effect among Asian students. Furthermore, 12th-grade math achieve-
ment showed a significant direct effect on STEM entrance. Also, it should be
noted that all three 12th-grade variables were significantly and positively
influenced by 10th-grade math achievement and attitudes, both of which
exerted significant indirect effects on STEM intent and STEM entrance.

1101
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Table 4

1102
Unstandardized and Standardized Estimates of Direct and Indirect Effects: Final Structural Equation Modeling (SEM) Model

White Asian URM

Model and Effect Unstandardized SE Standardized CPa Unstandardized SE Standardized CPa Unstandardized SE Standardized CPa

Direct effects
Math self-efficacy beliefs ON
10th-grade attitudes toward math 0.737*** .064 .377 (=) .420 (=) .437
10th-grade math achievement 0.073*** .007 .281 .044** .014 .232 .066*** .010 .319
Exposure to math and science ON
10th-grade attitudes toward math 0.175*** .018 .407 (=) .326 (=) .350
10th-grade math achievement 0.015*** .002 .271 (=) .270 (=) .252
12th-grade math achievement ON
10th-grade attitudes toward math 0.893*** .101 .118 (=) .098 (=) .112
10th-grade math achievement 0.810*** .010 .807 (=) .836 (=) .833
Intent to major in STEM ON
Math self-efficacy beliefs 0.101*** .021 .154 .034 (=) .153 .027 (=) .169 .036
Exposure to math and science 1.398*** .174 .468 .515 .838*** .235 .386 .287 .331** .118 .164 .124
12th-grade math achievement 0.029*** .006 .169 .010 .008 .008 .061 .016* .007 .127 .006
STEM entrance ON
Intent to major in STEM 1.120*** .072 .764 .331 (=) .701 .325 (=) .723 .410
12th-grade math achievement 0.017*** .005 .069 .003 (=) .083 .003 (=) .087 .007
Academic interaction 0.131** .050 .072 .024 (=) .092 .023 (=) .097 .051
College readiness in math and 0.167*** .051 .132 .032 (=) .294 .031 (=) .110 .065
science
Financial aid receipt 0.308** .102 .077 .043 (=) .074 .041 (=) .081 .111
Graduate degree expectation 0.385*** .100 .104 .051 (=) .101 .048 (=) .114 .137

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

Full-time enrollment –0.135 .285 –.013 (=) –.013 (=) –.020
Remedial reading –0.084 .170 –.016 (=) –.020 (=) –.018
Remedial writing –0.131 .160 –.030 (=) –.035 (=) –.033
Remedial math 0.070 .148 .016 (=) .018 (=) .019

(continued)
 Table 4 (continued)

White Asian URM

a a
Model and Effect Unstandardized SE Standardized CP Unstandardized SE Standardized CP Unstandardized SE Standardized CPa

Having dependent children –0.332* .163 –.050 –.045 (=) –.053 –.043 (=) –.059 –.119
Hours worked weekly –0.001 .006 –.005 (=) –.004 (=) –.005
Indirect effects
STEM Intent Efficacy 0.113*** .024 .118 (=) .107 (=) .024 .122
STEM Intent Exposure math/ 1.567*** .195 .358 .947*** .269 .273 .371** .134 .119
science
STEM Intent Math achievement 0.032*** .007 .129 .009 .009 .042 .018* .008 .092
(12th)
STEM Intent s Math attitudes 0.385*** .045 .205 .256*** .050 .138 .164*** .030 .105
(10th) (sum)
via s Efficacy 0.083*** .018 .044 (=) .045 (=) .018 .053
via s Exposure 0.274*** .038 .146 .165*** .049 .080 .065** .024 .041
via s Math achievement (12th) 0.029*** .007 .015 .008 .008 .004 .016* .007 .010
STEM Intent s Math 0.058*** .006 .234 .026*** .007 .134 .028*** .007 .145
achievement (10th) (sum)
via s Efficacy 0.008*** .002 .033 .005** .002 .025 .007*** .002 .039
via s Exposure 0.024*** .004 .097 .015*** .004 .074 .006** .002 .030
via s Math achievement (12th) 0.026*** .005 .104 .007 .007 .035 .015* .006 .077
Intent Efficacy Math attitudes 0.075*** .016 .058 (=) .064 (=) .074

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

(10th)
Intent Efficacy Math achieve- 0.007*** .001 .043 .004** .002 .036 .007*** .002 .054
ment (10th)
Intent Exposure Math attitudes 0.244*** .034 .191 .146*** .042 .126 .058** .021 .057
(10th)
Intent Exposure Math achieve- 0.022*** .003 .127 .013*** .004 .104 .005** .002 .041
ment (10th)

(continued)

1103
 Table 4 (continued)

1104
White Asian URM
a a
Model and Effect Unstandardized SE Standardized CP Unstandardized SE Standardized CP Unstandardized SE Standardized CPa

Intent Math achievement 0.026*** .006 .020 .007 .007 .006 .014* .006 .014
(12th) Math attitudes (10th)
Intent Math achievement (12th) 0.023*** .005 .136 .006 .006 .051 .013* .006 .106
Math achievement (10th)

Note. URM = underrepresented minorities; (=) = estimate constrained equal across groups. Two out of the five models depicted in this table have
a dichotomous outcome variable: intent to major in STEM (1 = intend to major in STEM, 0 = did not intend to major in STEM) and STEM entrance
(1 = entered into a STEM major, 0 = did not enter into a STEM major). For these two equations, probit models were conducted in Mplus. Although
the probit regression coefficients obtained from these analyses can show whether a particular independent variable has a positive or negative
effect on the probability that the dependent variable (e.g., intent to major in STEM) takes the value of 1, these coefficients are not as intuitive as
those of a linear regression. To translate the probit regression coefficients to probability values, the following formula (Muthén & Muthén, 1998–
2010) was adopted:

Pðy51jXÞ5F ða1b  XÞ5F ðt1b1  x1 1b2  x2 1 . . .Þ;

where F is the standard normal distribution function, a is the probit regression intercept, b is the probit regression coefficients, t is the probit
threshold, and t = –a. Building upon this formula, to present probit estimates more intuitively, Table 4 includes change in predicted probability
for each significant probit regression coefficient in the STEM intent and STEM entrance models. This value was obtained through the following
steps: First, compute P (y = 1 | x) with all continuous independent variables set to their mean values and discrete independent variables set to
their mode values to represent a typical case for a student. Then, to determine the impact of a unit change in a continuous independent variable
xj on P (y = 1), xj is set equal to x 11, and P (y = 1 | x) is recomputed with all other independent variables held at the same values as in the
typical case (i.e., other things being equal). The first probability value is then subtracted from the second probability value, and this difference is

Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016

the impact of a unit change in xj when all other independent variables are held at their mean or mode values. For the binary independent var-
iables in the models (e.g., financial aid receipt and expecting a graduate degree), the procedures are similar except that the binary xj variable is
set to 0 in the first step and then to 1 in the second step.
a
CP = change in probability (when outcome = 1), calculated with all other independent variables set to their mean (for continuous variables) or
mode (for discrete variables) value and reported for statistically significant direct effects in the intent to major in STEM and STEM entrance mod-
els only.
*p \ .05. **p \ .01. ***p \ .001.
 Why Students Choose STEM Majors

Postsecondary Context of Supports

Expecting a
Financial Enrollment
Graduate
Aid Intensity
Degree
Math
W: .737 A: (=) U: (=) Self-Efficacy Academic W: .308 W: .385 College
Beliefs A: (=) A: (=) Readiness in
Attitudes Interaction U: (=) U: (=)
12th Grade W: .101 Math and
Toward A: (=) W: .131 W: .167 Science
Math U: (=) A: (=) A: (=)
10th Grade U: (=) U: (=)
W: .175 A: (=) U: (=)
W: .893 A: (=) U: (=)
High School W: 1.398 Intent Entrance into a
W: 1.120
Exposure to A: .838 to Major in a A: (=) STEM
Math and U: .331
Science STEM Field U: (=) Field of Study
W: .073 A: .044 U: .066
W: .015 A: (=) U: (=)
Math W: .029 W: -.332
A: .008 A: (=)
Achievement
U: .016 W: .017 U: (=)
10th Grade
A: (=) Remediation External Demands
Math U: (=)
W: .810 A: (=) U: (=)
Achievement

Children
Reading

Writing

Having

Hours
Work
Math
12th Grade

Postsecondary Context of Barriers

2002 2004 2006

Figure 3. Results of final multiple-group structural equation modeling (SEM)

model based on race.
Note. W = White; A = Asian; U = underrepresented minorities; (=) = estimate was constrained
equal across groups. Insignificant paths are in gray.

Research Question 2: Taking into account the relationships described in Question

1, how are students’ initial postsecondary education experiences, such as aca-
demic interaction, receipt of financial aid, and remediation, related to STEM
entrance?

Intent to pursue STEM and several postsecondary latent and observed

variables showed direct effects on STEM entrance. Specifically, choosing
a STEM major was positively associated with intent to major in STEM, aca-
demic interaction, college readiness in math and science, receiving financial
aid, and expecting to earn a graduate degree. As for external demands, hav-
ing dependent children significantly and negatively affected STEM entrance
while number of weekly work hours did not have any significant effect.
Receiving remediation and being enrolled full-time did not show any influ-
ence on STEM entrance. None of these effects differ significantly across
racial, gender, and SES groups.

Research Question 3: How do the modeled effects vary based on gender, race, and
SES?

1105
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
The potentially varying effects of the modeled factors were examined
through conducting multiple-group SEM analyses based on race, gender,
and SES. These analyses indicated that the proposed theoretical model gen-
erally held well and was stable across various racial, gender, and SES groups.
Significant differences in structural weights were found in the multiple-
group model based on race. Specifically, the effect of 12th-grade math
achievement on intent to major in STEM was significant for White and under-
represented minority students, but was nonsignificant for Asian students. In
practical terms, for White students, a 1-point increase from the mean in math
achievement scores would result in a .010 increase in the probability of their
intending to major in STEM. For underrepresented minority students, this
change in the probability of STEM intent was .006 and for Asian American
students there would be no significant change.
While significantly affecting STEM intent of all students, exposure to math
and science had the largest impact on White students and the smallest effect
on underrepresented minority students. In practical terms, among White stu-
dents, when their exposure to math and science increased by 1 point above
the mean of this factor score, the increase in the probability that the students
would intend to major in STEM was .515. This increase in the probability of
STEM intent would be .287 for Asian students and only .124 for underrepre-
sented minority students. In addition, the overall significant and positive effect
of 10th-grade math achievement on 12th-grade math self-efficacy was most
substantial for underrepresented minorities, followed by White students and
Asian students. That is, a 1 standard deviation increase in 10th-grade math
achievement score was associated with .281 standard deviation increase in
the math self-efficacy factor score among White students, .232 standard devi-
ation increase among Asian students, and .319 standard deviation increase
among underrepresented minority students.

Discussion
As one of the first studies that applies the social cognitive career theory to
study a nationally representative high school cohort’s entrance into college
STEM majors, this research takes advantage of a unique, rich national data
set to holistically explore supports and barriers to STEM entrance. Results
point to important secondary and postsecondary factors influencing entrance
into STEM disciplines. In addition, pivotal racial differences are uncovered by
this study in terms of how early math-related attitudes and math and science
learning influence STEM choice. A closer examination of these results reveals
a number of important findings worthy of further discussion.

Math and Science Learning at the Secondary Level

High school preparation in math and science plays a critical role in
developing student interest in pursuing a STEM field of study and

1106
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
influencing entrance into STEM majors. Math and science learning in K–12
education has been central to the research and discussion on broadening
the STEM pipeline. In particular, selection and completion of math and sci-
ence courses during high school are essential in developing students’ predis-
positions toward choosing a STEM major in college (Blickenstaff, 2005).
The influence of high school learning as related to math and science,
however, is multifaceted. Many prior studies focused solely on math achieve-
ment when examining the influence of high school experience on student
interest and entrance into STEM fields (e.g., Crisp et al., 2009; Porter &
Umbach, 2006). This study, however, shows that the effect of students’ expo-
sure to math and science courses is even stronger than that of math achieve-
ment, which was once deemed the single best predictor of students’ future
STEM entrance. This finding implies that in order to boost high school stu-
dents’ interest in pursuing STEM fields of study, an earlier introduction
and exposure to math- and science-related courses could be an effective
method. This means that students’ interest in pursuing STEM could be an
evoked response to direct exposure to these courses.
On the other hand, math achievement still indicates a significant, persis-
tent effect on STEM intent (with the exception of Asian students whose
STEM goals and resulting persistence may originate prior to 12th grade7)
and subsequent enrollment in STEM majors. This warrants continued policy
focus on improving math achievement of students. When it comes to struc-
turing and engaging students in math and science courses, particular atten-
tion should be given to college readiness. As clearly indicated in this study,
students who perceive their high school math and science courses to have
adequately prepared them for college work are likely to choose a STEM
major. In light of these findings, a stronger alignment between high school
offerings and academic expectations at the college level represents a promis-
ing step toward promoting greater student interest and entrance into STEM
fields of postsecondary study.

The Importance of Motivational Beliefs

Motivation clearly matters in STEM-related interest and choices. The
study’s four motivational attributes, (a) attitudes toward math, (b) math
self-efficacy beliefs, (c) intent to pursue STEM fields of study, and (d) aspi-
ration to earn a graduate degree, all demonstrate a significant and positive
direct or indirect link to STEM entrance. From the pre-college perspective,
positive attitudes toward math (e.g., being interested in the subject and rec-
ognizing its importance) at an early age positively influence later math
achievement, math and science course-taking, and math self-efficacy beliefs,
all of which are the cornerstone of fostering intent to pursue STEM and even-
tually choosing these disciplines. While prior research has suggested that
positive attitudes toward math are fundamental to students’ persistence

1107
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
and success in math learning (Hackett & Betz, 1989; Singh, Granville, &
Dika, 2002), this study offers additional empirical evidence linking these at-
titudes with college students’ choice of STEM majors. The results illuminate
how these early attitudes affect STEM intent and entrance through their influ-
ence on 12th-grade factors that are critical for future STEM choice and suc-
cess, thus highlighting the importance of cultivating students’ positive
attitudes toward math from early on. In conjunction with prior research
(Bairaktarova & Evangelou, 2012; Marshall et al., 2011), this study’s findings
present viable approaches such as resorting to learning strategies that make
math education enjoyable and educating students about the significance and
long-term benefits of good math skills.
Math self-efficacy beliefs also play a significant and positive role in shap-
ing STEM intent, and through intent, math self-efficacy has a strong indirect
effect on actual STEM entrance. Similar to previous research that examined
the link between math self-efficacy and STEM choice (e.g., Scott &
Mallinckrodt, 2005), this study demonstrates that students with stronger
math self-efficacy beliefs are more likely to intend to major in STEM fields
upon college entrance. While this finding supports the argument for promot-
ing positive math self-efficacy beliefs among all students, it should be noted
that math self-efficacy often is discussed in relation to gender (Sadker &
Sadker, 1994). That is, male students are more self-efficacious in math
than female students despite comparable achievement (Eccles, 1994;
Pajares, 2005; Watt, 2006). Multiple-group analysis in this study shows that
there is no gender difference in terms of how math self-efficacy works to
influence students’ STEM intent. This suggests that improving female stu-
dents’ math self-efficacy may also help cultivate stronger interest in pursuing
STEM among female students with equivalent achievement in math as their
male counterparts. To make this happen, it is particularly important to fur-
ther address the gender bias in STEM discussion (Clewell & Campbell,
2002), which may adversely affect female students’ math self-efficacy beliefs.
In addition to self-efficacy, other key motivational factors in SCCT that
influence choice actions include outcome expectations and interests (Lent
et al., 2010). In this study, STEM intent is used as a proxy for outcome ex-
pectations and interests and is the biggest positive effect of all on the choice
action of interest: choosing a STEM major. This result aligns well with SCCT,
which stipulates that an individual’s intention to engage in a certain activity
(in this case choosing a major in STEM fields) helps organize, guide, and sus-
tain the individual’s efforts over a period of time.
This study also shows that expecting to earn a graduate degree is posi-
tively associated with STEM entrance. Perhaps those who are graduate
school aspirants tend to be a more select and motivated group who are suc-
cessful in establishing a stable, long-term academic plan and who are better
prepared to take on challenging fields of study such as STEM.

1108
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
Postsecondary Supports and Barriers
Postsecondary Supports
The first year of college is critical for students’ STEM choice, especially
when contextual supports in the form of interaction with faculty and aca-
demic advisors and receipt of financial aid are present. For all students, aca-
demic interaction seems to encourage entrance into STEM fields of study.
These interactions may help students better integrate themselves into the
college environment and also assist them in better aligning their academic
aspirations with actual choices. From an advising perspective, helping cur-
rent and potential STEM aspirants declare a STEM major early in their college
careers is critical to minimizing additional time, funds, and opportunity costs
spent in pursuing a degree (Frehill, 1997). As previously discussed, the
socialization process may help reinforce one’s academic and career choices.
Also, in the context of SCCT, such interactions serve as the contextual sup-
port that helps individuals persist in alignment with their goals.
Financial aid’s positive link to STEM entrance highlights the importance
of financial resources as another form of postsecondary support for students
pursuing STEM majors. It should be noted that given the correlational nature
of the analysis, this association can be interpreted in both directions: that
financial aid leads students to choose a STEM major or that students in
STEM majors are more likely to receive financial aid. Nonetheless, it seems
undeniable that financial resources provide the much needed support for
students to succeed in STEM fields of study. Students pursing STEM disci-
plines tend to spend more time studying than students in other fields
(Arum & Roksa, 2011; Brint, Cantwell, & Saxena, 2012). Therefore, receiving
financial aid may help relax financial constraints and allow them to allocate
enough time and energy to study and engage in greater interaction with fac-
ulty and advisors, thus meeting the academic challenges associated with ma-
joring in STEM fields. In fact, a recent study by Kienzl and Trent (2009)
showed that receiving financial aid helped socioeconomically underrepre-
sented students enter high-cost STEM fields at a large public research univer-
sity. Results from this research based on national data echo Kienzl’s and
Trent’s finding and further reveal that the positive effect of financial aid ap-
plies across racial, gender, and SES groups.

Postsecondary Barriers
In regard to variables representing postsecondary barriers, only being
a parent negatively affects STEM entrance, while remediation and number
of weekly work hours do not show any significant impact. Given that the
study’s sample reached their early 20s in 2006, students who entered parent-
hood by that time likely had children of a very young age. This presents con-
straints such as having demanding child care responsibilities and navigating

1109
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
a challenging schedule that may prevent them from enrolling in courses in
perfect alignment with their academic aspirations. These added challenges
may prevent students who are parents from choosing a STEM major.
The null effect of working hours may be due to the possible differential
effect of employment on college students’ academic experience and choices,
as evidenced in more recent studies on college student employment. It is
possible that students who work an ‘‘optimal’’ amount of hours and in
employment related to their academic interest gain skills (e.g., time manage-
ment and work-study balance) that help them make viable academic plans
and decisions (Dundes & Marx, 2006; Ehrenberg & Sherman, 1987). On
the other hand, working excessive hours and in areas isolated from one’s
academic work may put students at a disadvantage (Bean & Metzner,
1985). These potential varying effects of work hours may thus result in an
overall null effect of employment on STEM entrance. In a similar way, the
nonsignificant effect of remediation is likely due to the differential outcomes
of students’ remedial experience—with positive outcomes of taking remedi-
ation and negative ones cancelling each other out—that leads to an overall
null impact on STEM entrance.

Racial Differences Underlying STEM Entrance

For recent high school graduates, racial backgrounds still largely impact
the way in which high school math and science learning is linked to STEM
aspirations, with underrepresented minorities experiencing the least gain in
their intent to purse a STEM field through coursework exposure in math and
science. This result suggests that the effect of high school exposure to math
and science courses on STEM intent is heterogeneous, accruing more to
White students and least to underrepresented minorities.
The identification of this heterogeneous effect in regard to exposure to
math and science courses indicates that the well-documented racial disad-
vantage in STEM participation cannot simply be resolved by offering more
math and science to underrepresented minority students alone. Rather,
more research is needed to understand how math and science education
can better serve underrepresented minority students and what additional
factors contribute to these students’ STEM-related aspirations and choices.
One of the findings from this study suggests that math self-efficacy, a com-
mon positive influence on STEM intent, is influenced by early math achieve-
ment to a greater extent among underrepresented minority students than
among White and Asian students. This result is alarmingly important in
that if the current racial disparity in math achievement is not addressed early
enough, its adverse impact on future achievement of underrepresented
minorities will be compounded by its detrimental effect on math self-efficacy
beliefs. This will further discourage underrepresented minorities to pursue
and succeed in STEM fields. On the other hand, this finding also implies

1110
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
that interventions addressing math achievement of underrepresented racial
minorities should be implemented early on and if effective, may have large
impacts on these students’ math self-efficacy beliefs, thus promoting their
STEM interest and entrance. With these results, the current study pinpoints
the importance of paying attention to the potentially heterogeneous impacts
of various policies and practices when targeting underrepresented minorities
in expanding the STEM pipeline.

Pathway to STEM Majors: Holistic and Complex

Last but not least, the process leading to entrance into STEM fields of study
is complex; numerous influences—individual, psychological, contextual, and
social—act together to shape, develop, and sustain one’s interest and eventually
turn it into an actual choice. This study sets out to disentangle these influences
and their effects on STEM choice by adopting a holistic perspective that draws
upon the integration of SCCT and relevant literature on STEM education.
Although this study does not account for all of the complexities, its findings sug-
gest that SCCT is a viable framework for understanding STEM choice behaviors,
especially when secondary STEM-related learning experience is added to the
model. Consistent with SCCT and for all students, choosing a STEM major
largely is dependent on their intrinsic motivational attributes, such as math atti-
tudes, math self-efficacy beliefs, and interest in entering STEM. These observa-
tions, along with the finding that aspiring to earn a graduate degree turns out to
influence students’ STEM entrance, suggest that students’ pathways to STEM can
be substantially explained by their overall educational motivation and aspira-
tions. Students also respond to postsecondary supports and barriers, such as
academic interaction, receiving financial aid, and having dependent children,
when making choices in regard to STEM as a major field of study.
In addition, by utilizing multiple-group SEM analyses to gauge impact
heterogeneity, this study illuminates the differential processes leading up
to entrance into STEM majors among racial groups. An important finding
that emerges from this study is that the race-based inequitable participation
in STEM fields of study may be partially explained by the disparity in student
intent to major in STEM as a direct outcome of their pre-college learning ex-
periences in math and science: Although exposure to math and science
courses positively increases the likelihood of being interested in STEM ma-
jors for all students, this positive impact accrues most to White students and
least to underrepresented minority students.

Limitations of the Study and Directions for Future Research

The Study’s Limitations
This study’s findings should be considered in conjunction with several
important limitations. First, although the study relies on rich longitudinal

1111
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
data from a nationally representative sample, the use of an extant data set
poses conceptual and analytical constraints. One such constraint is the
time window covered by ELS:2002, which followed students from 10th grade
to 2 years after high school graduation. As delineated at various points in this
article, motivational beliefs, learning, and achievement in math formed ear-
lier in a child’s education have enduring effects on his or her future STEM-
related aspirations and choices. Although to the extent possible, the study
incorporates variables that speak to these early effects, similar variables
from the middle school or elementary school years are simply not available
given the design of ELS:2002. Therefore, relevant pre–high school influences
were not addressed in the study. Instead, variables from 10th grade were
used as proxies of earlier influences. Similarly, this study focuses on choos-
ing a STEM major roughly within 2 years of college. Some students may still
be exploring their major fields of interest during this time frame, and others
might switch into STEM disciplines later on. These dynamics were not
explored given the data available for the study. Therefore, it is impossible
to explore the long-term effects of the secondary and postsecondary varia-
bles on students’ entire progress through the STEM pipeline.
Also, variables of interest in existing data sets are not always measured
in ways the researcher would have desired. For example, self-efficacy beliefs
are central to SCCT, which serves as the guiding theoretical framework for
this study. While ELS:2002 contains survey items that measure math self-
efficacy adequately, multiple measures of STEM-related self-efficacy such
as science self-efficacy in addition to math self-efficacy would help disentan-
gle the complex nature of STEM learning. ELS does not include data for
developing such multiple measures, so this study relies on math self-efficacy
as a proxy for STEM self-efficacy beliefs, which limits the robustness of the
data in support of the theory.
On a similar note, intent to major in STEM fields of study is measured at
one point in time, gauging students’ interest in choosing a STEM major upon
entering postsecondary education. The one-time snapshot nature of this
measure limits the study’s ability to provide insights into how postsecondary
institutions may help develop students’ intent to choose a STEM field over
time. In addition, the lack of earlier measures of students’ intent to major
in STEM prohibits the study from assessing the potential impact of these ear-
lier aspirations for a STEM college career on high school students’ attitude
toward math and math achievement.
Another limitation relates to the lack of causal inference, given the use of
extant survey data and SEM. ELS:2002 provides observational data that did
not involve any random assignment of students to any of the independent
variables, such as financial aid receipt or high school exposure to math
and science. Although SEM goes beyond the traditional regression analysis
in that it accounts for the temporal, complex relationships among latent
and observed variables, it is still an exploration of various correlations.

1112
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
While plausible explanations for the findings are discussed based on theory
and prior research, none of the relationships described in this article should
be interpreted as causal.

Directions for Future Research

This study points to several directions for future research. Although
studies on STEM-related issues have proliferated in the past decade, disen-
tangling the reasons for STEM participation remains a complex challenge,
especially when determining whether the factors are dependent on sociode-
mographic backgrounds and/or within the control of the educator.
Integrating variables at the secondary and postsecondary levels, this study
accounts for a longitudinal process of STEM choice and tackles effect hetero-
geneity based on race, gender, and SES. However, a number of questions
remain to be answered in future research. First, the finding surrounding
math self-efficacy needs further investigation to understand more completely
the mechanism through which it works in support of interest in STEM fields
and future STEM choice, especially among female students. How does math
self-efficacy or, better yet, STEM self-efficacy interact with various socializa-
tion sources and social perceptions regarding the gender role in career- and
major-related choices to affect students’ actual STEM choices? Which one of
the sources of self-efficacy is the most theoretically and practically viable
way to help promote STEM-related self-efficacy: mastery experience, vicari-
ous experience, social persuasions, or physiological factors?
Furthermore, given the persistent, enduring effect of high school expo-
sure to math and science courses, as well as math achievement, finding the
best possible way to teach those courses, especially accounting for racial dif-
ferences in the ways in which these effects are transmitted, will continue to
dominate the central stage of STEM discussion. Also important to note, rigor-
ously designed experimental or quasi-experimental studies may represent the
best approach for scaling up promising interventions: An experimental design,
by randomly assigning students to an intervention, can establish the true effec-
tiveness of a program aimed at promoting STEM participation. Alternatively,
quasi-experimental research, often by adopting a rigorous approach to creat-
ing comparison groups, can also estimate (although not as reliably as true ex-
periments can) causal effects of a STEM intervention or practice.
Equally important, the first year in college can be critical, and as evi-
denced in this study, a number of postsecondary variables are related to
entrance into STEM fields of study. Of particular note are the positive effects
of both receiving financial aid and academic interaction on STEM entrance.
These relationships need to be further studied—ideally through original, tar-
geted data collection—to understand how they affect STEM entrance. Such
nuanced understanding may aid in the development of policy interventions
that truly can make a difference.

1113
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Conclusion
This study addresses the vital secondary-postsecondary nexus in STEM
entrance, an issue often reflected in policy discussions but seldom systemat-
ically examined in empirical research, especially from a longitudinal
approach. Given the rising national attention to promoting seamless move-
ment through the STEM pipeline among students of diverse backgrounds,
continued policy focus will be given to participation of traditionally under-
represented groups. To support this policy priority, a comprehensive knowl-
edge of the barriers and facilitators to entering these fields of study is of
paramount importance.
Following a holistic view of the issue of inequity in STEM participation,
this study uncovers the impact of critical motivational, secondary learning,
and postsecondary variables on STEM entrance and establishes the social
cognitive career theory as a viable conceptual model for future STEM-related
research. Furthermore, results from this study illuminate important racial dif-
ferences in how pre-college learning and motivation exert their influence on
students’ intent to major in STEM. In light of these findings, educational pol-
icy and interventions aimed at developing STEM-related perceptions, atti-
tudes, and aspirations among underrepresented minority students will
benefit from a deeper understanding of the potentially heterogeneous effects
of variable educational experiences. Together, results from this study offer
new theoretical and empirical knowledge that informs policy and practice
intended to promote equitable participation in STEM fields of postsecondary
study.

Notes
This study is based upon work supported by the Association for Institutional
Research, the National Center for Education Statistics, the National Science Foundation,
and the National Postsecondary Education Cooperative under Association for
Institutional Research Grant Number RG11-07. Any opinions, findings, and conclusions
or recommendations expressed in this material are those of the author and do not neces-
sarily reflect the views of the Association for Institutional Research, the National Center for
Education Statistics, the National Science Foundation, or the National Postsecondary
Education Cooperative. Nik Hawkins, Sarah Hurley, Kelly Wickersham, and several anon-
ymous reviewers provided helpful comments on earlier drafts of the article.
1
Following the suggestion made by one of the reviewers, two sets of high school var-
iables were also analyzed as additional covariates in the structural equation modeling
(SEM) model: (a) family background including first-generation status (1 = first-generation
college student; 0 = continuing generation) and language background (1 = English is
native language; 0 = English is not native language) and (b) high school context variables
including percentage of the school’s students that qualify for free/reduced-price lunch,
percentage of minority students in the school, student-teacher ratio of the school, high
school type (dummy coded into Catholic, other private, and public as the referent group),
and urbanicity of the school (dummy coded into suburban, rural, and urban as the refer-
ent group). Analysis of these added covariates indicated that none of them had a statisti-
cally significant relationship to STEM entrance and the effects of variables already in the
model did not change substantially.

1114
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
2
The mathematics test standardized score was a T-score created by a transformation
of the IRT (item response theory) theta (ability) estimate, rescaled to a mean of 50 and
standard deviation of 10, from the cognitive assessments in Education Longitudinal
Study of 2002 (ELS:2002). The standardized T-score provides a norm-referenced estimate
of achievement relative to the population (spring 2002 10th graders and spring 2004 12th
graders, respectively) as a whole (Source: ELS:2002 Electronic Codebook).
3
After the full analysis, a series of interaction terms were added to this final regression
equation. These interaction terms were created between intent to major in a STEM field
and variables indicating postsecondary context of supports of barriers to assess the poten-
tial interaction effects between intent and postsecondary context variables. Results
showed that none of the interaction terms was statistically significant.
4
In Mplus, the residual covariance between math achievement and math self-efficacy
at 12th grade was estimated by adding the ‘‘WITH’’ statement between these two variables.
5
Underrepresented minorities include African Americans, Hispanics, Native
Americans, and multiracial students. In the literature highlighting inequitable participation
in STEM education by race, three key racial and ethnic groups, African Americans,
Hispanics, and Native Americans, are often analyzed together in comparison to their
White and Asian counterparts. STEM-related research and data on students who identify
their race/ethnicity as ‘‘multiracial’’ are scarce and in this sense they are also underrepre-
sented. Also, if each race category were to represent a distinct group, the multiple-group
SEM analysis would become challenging to conduct and interpret. In addition, the small
numbers of Native American and multiracial students make it difficult to analyze them sep-
arately. Given these theoretical and analytical considerations, these racial/ethnic groups
were combined as the underrepresented minorities in STEM.
6
As a regular practice, the chi-square value is almost always presented in studies that
involve confirmatory factor analysis (CFA) and SEM (Kline, 2011). However, because the
chi-square test is sensitive to sample size (Kenny, 2011; Schumacker & Lomax, 2004), it
might erroneously suggest a poor fit by rejecting the null hypothesis in studies with large
sample sizes like this. As a result, other fit indices such as Comparative Fit Index (CFI) and
Tucker-Lewis Fit Index (TLI) are more relevant to this study. Also reported is the relative
chi-square, alternatively referred to as the normed chi-square, which equals the chi-square
value divided by the degrees of freedom. Some scholars argue that this index might be less
sensitive to sample size, but the guidelines about acceptable maximum values vary, rang-
ing from less than 2 (e.g., Ullman, 2001) to less than 5 (e.g., Schumacker & Lomax, 2004).
7
For Asian American students, 12th-grade math achievement did not emerge as a sig-
nificant factor associated with their STEM intent. Asian Americans are well represented in
STEM fields (Anderson & Kim, 2006; May & Chubin, 2003), and this high representation
may well be a result of Asian students’ stronger aspirations to pursue math- and
science-related careers at a very young age, a level unmatched by any other racial groups
(National Science Foundation, 1994). This early interest, although not accounted for in this
study given the limitation of the data, may have largely translated into their STEM intent
independent of their 12th-grade math achievement.

References
ACT. (2006). Developing the STEM education pipeline. Iowa City, IA: Author.
Adelman, C. (1998). Answers in the tool box: Academic intensity, attendance pat-
terns, and bachelor’s degree attainment. Retrieved from http://www.ed.gov/
pubs/Toolbox/Title.html
Adelman, C. (1999). Women and men of the engineering path: A model for analysis
of undergraduate careers. Washington, DC: U.S. Department of Education.
Adelman, C. (2006). The toolbox revisited: Paths to degree completion from high
school through college. Washington, DC: U.S. Department of Education.
Anderson, E., & Kim, D. (2006). Increasing the success of minority students in science
and technology. Washington, DC: American Council on Education.

1115
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Arum, R., & Roksa, J. (2011). Academically adrift: Limited learning on college cam-
puses. Chicago, IL: University of Chicago Press.
Astin, A. W. (1993). What matters in college? Four critical years revisited. San
Francisco, CA: Jossey-Bass.
Bahr, P. R. (2008). Does mathematics remediation work?: A comparative analysis of
academic attainment among community college students. Research in Higher
Education, 49(5), 420–450.
Bailey, T., & Alfonso, M. (2005). Paths to persistence: An analysis of research on pro-
gram effectiveness at community colleges (New Agenda Series Vol. 6, No. 1).
Indianapolis, IN: Lumina Foundation for Education. Retrieved from http://
www.luminafoundation.org/publications/PathstoPersistence.pdf
Bailyn, L. (2003). Academic careers and gender equity: Lessons learned from MIT.
Gender, Work & Organization, 10(2), 137–153.
Bairaktarova, D., & Evangelou, D. (2012). Creativity and science, technology, engi-
neering, and mathematics (STEM) in early childhood education. In
O. N. Saracho (Ed.), Contemporary perspectives on research in creativity in early
childhood education (pp. 377–396). Charlotte, NC: Information Age Publishing.
Bandura, A. (1986). Social foundations of thought and action: A social cognitive the-
ory. Englewood Cliffs, NJ: Prentice Hall.
Bean, J. P., & Metzner, B. S. (1985). A conceptual model of nontraditional undergrad-
uate student attrition. Review of Educational Research, 55(4), 485–540.
Berkner, L. K., Cuccaro-Alamin, S., & McCormick, A. C. (1996). Descriptive summary
of 1989–90 beginning postsecondary students: Five years later with an essay on
postsecondary persistence and attainment (NCES 96-155). Washington, DC:
National Center for Education Statistics.
Betz, N. E., & Hackett, G. (1983). The relationship of mathematics self-efficacy ex-
pectations to the selection of science-based college majors. Journal of
Vocational Behavior, 23(3), 329–345.
Blickenstaff, J. C. (2005). Women and science careers: Leaky pipeline or gender filter.
Gender and Education, 17(4), 369–386.
Bowman, B. T. (1998). Math, science and technology in early childhood education.
Paper presented at the forum on Early Childhood Science, Mathematics, and
Technology Education, Washington, DC.
Brainard, S., Metz, S. S., & Gillmore, G. (1993). WEPAN pilot climate survey: Exploring
the environment for undergraduate engineering students. Retrieved from http://
www.wepan.org/associations/5413/files/Climate%20Survey.pdf
Brint, S., Cantwell, A., & Saxena, P. (2012). Disciplinary categories, majors, and
undergraduate academic experiences: Rethinking Bok’s ‘‘underachieving col-
leges’’ thesis. Research in Higher Education, 53(1), 1–25.
Byars-Winston, A. M., Estrada, Y. E., Howard, C. H., Davis, D., & Zalapa, J. (2010).
Influence of social cognitive and ethnic variables on academic goals of under-
represented students in science and engineering. Journal of Counseling
Psychology, 57(2), 205–218.
Byrne, B. M. (1998). Structural equation modeling with LISREL, PRELIS, and SIMPLIS:
Basic concepts, applications, and programming. Mahwah, NJ: Lawrence
Erlbaum.
Byrne, B. M. (2010). Structural equation modeling with AMOS: Basic concepts, appli-
cations, and programming (2nd ed.). New York, NY: Routledge.
Cabrera, A. F., Colbeck, C. L., & Terenzini, P. T. (2001). Developing performance in-
dicators for assessing classroom teaching practices and student learning: The
case of engineering. Research in Higher Education, 42(3), 327–352.

1116
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
Carter, D. F. (2002). College students’ degree aspirations: A theoretical model and lit-
erature review with a focus on African American and Latino students. In
J. C. Smart (Ed.), Higher education: A handbook of theory and research (pp.
129–171). Bronx, NY: Agathon Press.
Carter, D. F. (2006). Key issues in the persistence of underrepresented minority stu-
dents. New Directions for Institutional Research, 130, 33–46.
Chang, J. C. (2005). Faculty-student interaction at the community college: A focus on
students of color. Research in Higher Education, 46(7), 769–802.
Chang, M. J., Cerna, O., Han, J., & Sáenz, V. (2008). The contradictory roles of insti-
tutional status in retaining underrepresented minorities in biomedical and
behavioral science majors. The Review of Higher Education, 31(4), 433–464.
Chang, M. J., Sharkness, J., Newman, C., & Hurtado, S. (2010, May). What matters in
college for retaining aspiring scientists and engineers? Paper presented at the
annual meeting of the American Educational Research Association, Denver, CO.
Chen, X., & Weko, T. (2009). Students who study science, technology, engineering,
and mathematics (STEM) in postsecondary education (NCES 2009-61).
Washington, DC: National Center for Education Statistics.
Clewell, B. C., & Campbell, P. B. (2002). Taking stock: Where we’ve been, where we
are, where we’re going. Journal of Women and Minorities in Science and
Engineering, 8, 255–284.
Crisp, G., Nora, A., & Taggart, A. (2009). Student characteristics, pre-college, college,
and environmental factors as predictors of majoring in and earning a STEM
degree: An analysis of students attending a Hispanic Serving Institution.
American Educational Research Journal, 46(4), 924–942.
DesJardins, S., Ahlburg, D., & McCall, B. P. (2006). An integrated model of applica-
tion, admission, enrollment, and financial aid. Journal of Higher Education,
77(3), 381–429.
Dowd, A. C., Malcom, L. E., & Bensimon, E. M. (2009). Benchmarking the success of
Latina and Latino students in STEM to achieve national graduation goals. Los
Angeles, CA: Center for Urban Education.
Dundes, L., & Marx, J. (2006). Balancing work and academics in college: Why do stu-
dents working 10-19 hours per week excel? Journal of College Student Retention,
8(1), 107–120.
Eagan, M. K. (2009, November). An examination of the contributors to production
efficiency of undergraduate degrees in STEM. Paper presented at the annual
meeting of the Association for the Study of Higher Education, Vancouver, BC,
Canada.
Eccles, J. S. (1994). Understanding women’s educational and occupational choices:
Applying the Eccles et al. model of achievement-related choices. Psychology of
Women Quarterly, 18(4), 585–609.
Ehrenberg, E. G., & Sherman, D. R. (1987). Employment while in college, academic
achievement, and postcollege outcomes: A summary of results. Journal of
Human Resources, 22(1), 1–23.
Ellington, R. M. (2006). Having their say: Eight high-achieving African-American
undergraduate mathematics majors discuss their success and persistence in
mathematics (Unpublished doctoral dissertation). University of Maryland,
College Park. Retrieved from Dissertations and Theses Database.
Elliott, R., Strenta, A. C., Adair, R., Matier, M., & Scott, J. (1996). The role of ethnicity in
choosing and leaving science in highly selective institutions. Research in Higher
Education, 37(6), 681–709.
Ethington, C. A., & Wolfle, L. M. (1988). Women’s selection of quantitative undergrad-
uate fields of study. American Educational Research Journal, 25(2), 157–175.

1117
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Fox, M. A., & Hackerman, N. (Eds.). (2003). Evaluating and improving undergrad-
uate teaching in science, technology, engineering, and mathematics.
Washington, DC: National Academies Press.
Frehill, L. M. (1997). Education and occupational sex segregation: The decision to
major in engineering. Sociological Quarterly, 38(2), 225–249.
Goldrick-Rab, S., Carter, D. F., & Winkle-Wagner, R. (2007). What higher education
has to say about the transition to college. Teachers College Record, 109(10),
2444–2481.
Gruca, J. M., Ethington, C. A., & Pascarella, E. T. (1988). Intergenerational effects of
college graduation on career sex atypicality in women. Research in Higher
Education, 29(2), 99–124.
Hackett, G., & Betz, N. E. (1989). An exploration of the mathematics self-efficacy/
mathematics performance correspondence. Journal for Research in Mathematics
Education, 20(3), 261–273.
Hackett, G., Betz, N. E., Casas, J. M., & Rocha-Singh, I. A. (1992). Gender, ethnicity,
and social cognitive factors predicting the academic achievement of students in
engineering. Journal of Counseling Psychology, 39(4), 527–538.
Herrera, F. A., & Hurtado, S. (2011). Maintaining initial interests: Developing science,
technology, engineering, and mathematics (STEM) career aspirations among
underrepresented racial minority students. Paper presented at the Association
for Educational Research Annual Meeting, New Orleans, LA.
Holland, J. (1992). Making vocational choices: A theory of vocational personalities
and work environments. Lutz, FL: Psychological Assessment Resources.
Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modeling:
Guidelines for determining model fit. Electronic Journal of Business Research
Methods, 6(1), 53–60.
Huang, G., Taddese, N., & Walter, E. (2000). Entry and persistence of women and
minorities in college science and engineering education (NCES 2000-601).
Washington, DC: National Center for Education Statistics.
Ishitani, T. T., & DesJardins, S. L. (2002). A longitudinal investigation of dropout from
colleges in the United States. Journal of College Student Retention, 4(2), 173–201.
Kaplan, D. (2009). Structural equation modeling: Foundations and extensions (2nd
ed.). Thousand Oaks, CA: Sage.
Kenny, D. A. (2011). Measuring model fit. Retrieved from http://davidakenny.net/
cm/fit.htm
Kienzl, G., & Trent, W. (2009). Underrepresented undergraduates in STEM: From
matriculation to degree completion at large, research-intensive public universi-
ties. Report to the Alfred P. Sloan Foundation, New York, NY.
Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed.).
New York, NY: Guilford Press.
Kulis, S., & Sicotte, D. (2002). Women scientists in academia: Geographically con-
strained to big cities, college clusters, or the coasts? Research in Higher
Education, 43(1), 1–30.
Kupek, E. (2006). Beyond logistic regression: Structural equations modeling for
binary variables and its application to investigating unobserved confounders.
BMC Medical Research Methodology, 6(1), 13–23.
Lacey, T. A., & Wright, B. (2009). Occupational employment projections to 2018.
Monthly Labor Review, 132(11), 82–123.
Leach, E. (2010). Challenges and opportunities for increasing the STEM pipeline. The
League for Innovation in the Community College, 23(9), 1.

1118
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
Lent, R. W., & Brown, S. D. (2006). Integrating person and situation perspectives on
work satisfaction: A social-cognitive view. Journal of Vocational Behavior,
69(2), 236–247.
Lent, R. W., Brown, S. D., & Hackett, G. (1994). Toward a unifying social cognitive
theory of career and academic interest, choice and performance. Journal of
Vocational Behavior, 45(1), 79–122.
Lent, R. W., Brown, S. D., & Hackett, G. (2000). Contextual supports and barriers to
career choice: A social cognitive analysis. Journal of Counseling Psychology,
47(1), 36–49.
Lent, R. W., Lopez, A. M., Lopez, F. G., & Sheu, H. (2008). Social cognitive career the-
ory and the prediction of interests and choice goals in the computing disciplines.
Journal of Vocational Behavior, 73(1), 52–62.
Lent, R. W., Lopez, F. G., & Bieschke, K. J. (1993). Predicting mathematics-related
choice and success behaviors: Test of an expanded social cognitive model.
Journal of Vocational Behavior, 42(2), 223–236.
Lent, R. W., Sheu, H., Gloster, C. S., & Wilkins, G. (2010). Longitudinal test of the
social cognitive model of choice in engineering students at historically Black uni-
versities. Journal of Vocational Behavior, 76(3), 387–394.
Long, B. T. (2005) The remediation debate: Are we serving the needs of underpre-
pared college students? National CrossTalk, 13(4), 11–12.
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and
determination of sample size for covariance structure modeling. Psychological
Methods, 1(2), 130–149.
Maple, S. A., & Stage, F. K. (1991). Influences on the choice of math/science major by
gender and ethnicity. American Educational Research Journal, 28(1), 37–60.
Marshall, S. P., McGee, G. W., McLaren, E., & Veal, C. C. (2011). Discovering and
developing diverse STEM talent: Enabling academically talented urban youth
to flourish. Gifted Child Today, 34(1), 16–23.
May, G., & Chubin, D. (2003). A retrospective on undergraduate engineering success
for underrepresented minorities. Journal of Engineering Education, 92(1), 27–
40.
Muthén, L. K., & Muthén, B. O. (1998–2010). Mplus user’s guide (6th ed.). Los
Angeles, CA: Muthén & Muthén.
National Academies 2005 ‘‘Rising Above the Gathering Storm’’ Committee. (2010).
Rising above the gathering storm: Rapidly approaching Category 5.
Washington, DC: Author.
National Science Board. (2004). Science and engineering indicators 2004.
Washington, DC: U.S. Government Printing Office.
National Science Board. (2010). Preparing the next generation of STEM innovators:
Identifying and developing our nation’s human capital. Arlington, VA: Author.
National Science Foundation. (1994). Women, minorities, and persons with disabil-
ities in science and engineering: 1994 (NSF 94-333). Arlington, VA: Author.
National Science Foundation. (2006). Science and engineering indicators 2006.
Arlington, VA: Author.
National Science Foundation. (2010). Science and engineering indicators 2010.
Arlington, VA: Author.
Olitsky, N. H. (2012). How do academic achievement and gender affect the earnings
of STEM majors? A propensity score matching approach. Retrieved from http://
www.econ.uconn.edu/seminars/20122013/papers/STEM_Olitsky.pdf
Pajares, F. (1996). Self-efficacy beliefs and mathematical problem-solving of gifted
students. Contemporary Educational Psychology, 21(4), 325–344.

1119
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
Wang
Pajares, F. (2005). Gender differences in mathematics self-efficacy beliefs. In
A. M. Gallagher & J. C. Kaufman (Eds.), Mind the gap: Gender differences in
mathematics (pp. 294–315). Boston, MA: Cambridge University Press.
Pajares, F., & Kranzler, J. H. (1995). Self-efficacy beliefs and general mental ability in
mathematical problem-solving. Contemporary Educational Psychology, 20(4),
426–443.
Pajares, F., & Miller, M. D. (1994). Role of self-efficacy and self-concept beliefs in
mathematical problem-solving: A path analysis. Journal of Educational
Psychology, 86(2), 193–203.
Pascarella, E. T., & Terenzini, P. T. (2005). How college affects students: Vol. 2. A third
decade of research. San Francisco, CA: Jossey-Bass.
Porter, S. R., & Umbach, P. D. (2006). College major choice: An analysis of person-
environment fit. Research in Higher Education, 47(4), 429–449.
Russell, M. L., & Atwater, M. M. (2005). Traveling the road to success: A discourse on
persistence throughout the science pipeline with African American students at
a predominantly White institution. Journal of Research in Science Teaching,
42(6), 691–715.
Sadker, M., & Sadker, D. (1994). Failing at fairness: How America’s schools cheat
girls. New York, NY: Simon & Schuster.
Schreiber, J. B., Stage, F. K., King, J., Nora, A., & Barlow, E. A. (2006). Reporting struc-
tural equation modeling and confirmatory factor analysis results: A review.
Journal of Educational Research, 99(6), 323–337.
Schultz, P. W., Hernandez, P. R., Woodcock, A., Estrada, M., Chance, R. C., Aguilar,
M., & Serpe, R. T. (2011). Patching the pipeline: Reducing educational disparities
in the sciences through minority training programs. Educational Evaluation and
Policy Analysis, 33(1), 95–114.
Schumacker, R. E., & Lomax, R. G. (2004). A beginner’s guide to structural equation
modeling (2nd ed.). Mahwah, NJ: Lawrence Erlbaum Associates.
Scott, A. B., & Mallinckrodt, B. (2005). Parental emotional support, science self-
efficacy, and choice of science major in undergraduate women. The Career
Development Quarterly, 53(3), 263–273.
Seymour, E., & Hewitt, N. (1997). Talking about leaving: Why undergraduates leave
the sciences. Boulder, CO: Westview Press.
Singh, K., Granville, M., & Dika, S. (2002). Mathematics and science achievement:
Effects of motivation, interest, and academic engagement. Journal of
Educational Research, 95(6), 323–332.
Strayhorn, T. L. (2010, October). Work in progress—Social barriers and supports to
underrepresented minorities’ success in STEM fields. Paper presented at
Frontiers in Education Conference, Washington, DC.
Terenzini, P. T., Pascarella, E. T., & Blimling, G. S. (1999). Students’ out-of-class ex-
periences and their influence on learning and cognitive development: A litera-
ture review. Journal of College Student Development, 40(5), 610–622.
Toker, Y., & Ackerman, P. L. (2012). Utilizing occupational complexity levels in voca-
tional interest assessments: Assessing interests for STEM areas. Journal of
Vocational Behavior, 80(2), 524–544.
Trusty, J. (2002). Effects of high school course-taking and other variables on choice of
science and mathematics college majors. Journal of Counseling and
Development, 80(4), 464–474.
Tyson, W., Lee, R., Borman, K. M., & Hanson, M. A. (2007). Science, technology, engi-
neering, and mathematics (STEM) pathways: High school science and math
coursework and postsecondary degree attainment. Journal of Education for
Students Placed at Risk, 12(3), 243–270.

1120
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016
 Why Students Choose STEM Majors
Ullman, J. B. (2001). Structural equation modeling. In B. G. Tabachnick & L. S. Fidell
(Eds.), Using multivariate statistics (pp. 653–771). Needham Heights, MA: Allyn
& Bacon.
U.S. Department of Labor. (2007). The STEM workforce challenge: The role of the pub-
lic workforce system in a national solution for a competitive science, technology,
engineering, and mathematics (STEM) workforce. Washington, DC: Author.
Wang, X. (2009). Baccalaureate attainment and college persistence of community col-
lege transfer students at four-year institutions. Research in Higher Education,
50(6), 570–588.
Wang, X. (2013). Baccalaureate expectations of community college students: Socio-
demographic, motivational, and contextual Influences. Teachers College
Record, 115(4). Retreived from http://www.tcrecord.org/library
Ware, N. C., & Lee, V. E. (1988). Sex differences in choice of college science majors.
American Educational Research Journal, 25(4), 593–614.
Watt, H. M. G. (2006). The role of motivation in gendered educational and occupa-
tional trajectories related to maths. Educational Research and Evaluation,
12(4), 305–322.

Manuscript received February 20, 2012

Final revision received February 28, 2013
Accepted April 8, 2013

1121
Downloaded from http://aerj.aera.net at CORNELL UNIV on October 28, 2016