JOURNAL OF RESEARCH IN SCIENCE TEACHING
VOL. 51, NO. 5, PP. 555–580 (2014)
Research Article
What Matters in College for Retaining Aspiring Scientists and Engineers
From Underrepresented Racial Groups
Mitchell J. Chang,1 Jessica Sharkness,2 Sylvia Hurtado,1 and Christopher B. Newman3
1
2
Department of Education, University of California, Los Angeles, California 90095-1521
Office of Institutional Research and Evaluation, Tufts University, Medford, Massachusetts
3
Department of Leadership Studies, University of San Diego, San Diego, California
Received 30 April 2012; Accepted 16 January 2014
Abstract: This longitudinal study examined factors that contribute to the persistence of underrepresented racial minority (URM) undergraduates in STEM fields. The primary source of data came from the
Cooperative Institutional Research Program’s 2004 The Freshman Survey (TFS) and 2008 College Senior
Survey (CSS). The sample included 3,670 students at 217 institutions who indicated on the TFS that they
intended to major in a STEM field, 1,634 of whom were underrepresented minority (URM) students. Findings
indicate that Black and Latino undergraduates were significantly less likely to persist in STEM majors than
were their White and Asian American counterparts. Background characteristics and college experiences
moderated this race effect, suggesting both that pre-college factors may explain some of the observed racial
disparities and that individual institutions can take more concrete actions to improve science achievement.
Findings from the follow-up analysis of the sample of URMs suggest that institutions can improve URM
STEM persistence by increasing the likelihood that those students will engage in key academic experiences:
studying frequently with others, participating in undergraduate research, and involvement in academic clubs
or organizations. # 2014 Wiley Periodicals, Inc. J Res Sci Teach 51: 555–580, 2014
Keywords: race; underrepresented racial minorities; achievement gap; persistence; retention; student
experience
For nearly a decade, governmental agencies (e.g., AAAS, 2001 and National Academy of
Sciences, National Academy of Engineering, & Institute of Medicine, 2007) have maintained that
the productivity and strength of the U.S. economy will face a serious decline if no significant
action is taken to address the racial disparities in the attainment of post-secondary degrees in
science, technology, engineering, and mathematics (STEM) fields. The National Science
Foundation (NSF, 2009) reported that although African American and Latino undergraduates
were just as likely as their White counterparts to enter college with the intention to major in
STEM, they were much less likely to earn a degree in those majors. A National Research Council
(NRC, 2011) report, Expanding Underrepresented Minority Participation: America’s Science
and Technology Talent at the Crossroads, stated that most of the growth in the new jobs will require
science and technology skills, and that “those groups that are most underrepresented in S&E are
also the fastest growing in the general population” (p. 3). Indeed, they write, the proportion of
Contract grant sponsor: National Institute of General Medical Sciences; Contract grant numbers: R01 GMO7196801; Contract grant sponsor: National Science Foundation; Contract grant number: 0757076.
Correspondence to: Mitchell J. Chang; E-mail: [email protected]
DOI 10.1002/tea.21146
Published online 18 February 2014 in Wiley Online Library (wileyonlinelibrary.com).
# 2014 Wiley Periodicals, Inc.
556
CHANG ET AL.
underrepresented minorities in science and engineering would need to triple to match their share
in the population. In an effort to achieve long-term parity in the preparation of a diverse workforce,
the NAS recommends a near-term goal of doubling the number of underrepresented minorities
receiving undergraduate STEM degrees.
As suggested by these reports, the underrepresentation of racial/ethnic minorities is not
necessarily attributable to a lack of interest in science fields, but rather to poor degree completion
rates. The latter pattern has been well documented. For example, Huang, Taddese, and Walter
(2000) found that African American, Latino, and Native American students—or underrepresented
racial minorities (URMs)—had lower persistence rates in science and engineering (26%) than
their White and Asian American counterparts (46%). A more recent study conducted by the
Higher Education Research Institute (2010) found that 33% of White and 42% of Asian American
students at a national sample of institutions completed their bachelor’s degree in STEM within
5 years of entering college, compared to only 18% of African American and 22% of Latino
students. URMs, then, appear to face unique and persistent challenges when moving along the
STEM pipeline. However, according to Mutegi (2013), science education research has been
unable to explain how science degree achievement is “racially determined.” Although students’
ability and pre-college preparation are important, steady progress through the STEM pipeline also
depends on the types of opportunities, experiences, and support students receive while in college
(Chang, Eagan, Lin, & Hurtado, 2011; Espinosa, 2011).
Institutions of higher education stand to play an important role in improving the participation
rates of key segments of our nation’s population in STEM-related careers and subsequently, also
contribute to our nation’s long-term welfare. Perhaps the most immediate and obvious role that
they can play is to do a better job retaining the African American, Native American, and Latino
undergraduates who enter college seeking a degree in a STEM field. In the interest of achieving
that goal, the main purpose of this study was to identify key individual and institutional factors that
either positively or negatively predict STEM degree persistence. We first examined whether or not
a student’s racial classification contributed to the chances of persisting in a STEM major after
4 years of college, and then examined the extent to which other factors moderated this race effect.
To the extent that the effects of race can be explained by other factors, we examined patterns of
preparation and experiences that contributed to the persistence of URM students.
Literature Review
Previous studies have identified a number of factors that contribute to the retention of
undergraduates in 4-year colleges and universities in general (see for example Nora, Barlow, &
Crisp, 2005; Tinto, 1993; Titus, 2004, 2006) and STEM fields in particular (Daempfle, 2003;
Elliott, Strenta, Adair, Matier, & Scott, 1996; Grandy, 1998; Seymour & Hewitt, 1997; White,
Altschuld, & Lee, 2006). Some of the factors that have been shown to contribute to STEM
retention include students’ high school academic preparation (Elliott et al., 1996), freshmen
academic performance and enrollment in “gatekeeper” courses (Crisp, Nora, & Taggart, 2009),
personal contact with faculty (Daempfle, 2003), degree of racial isolation (Seymour &
Hewitt, 1997), exposure to racial minority support systems (Grandy, 1998), and an institution’s
level of selectivity (Chang, Cerna, Han, & Sáenz, 2008; Chang et al., 2011; Espinosa, 2011).
Taken together, these studies suggest that completion of a STEM degree requires not only
academic preparation but also resilience and capacity to negotiate a complex academic context.
In Seymour and Hewitt’s (1997) seminal qualitative study of 425 undergraduates who entered
seven universities as STEM majors, the experiences of students who changed majors were
compared to the experiences of those who persisted. The researchers found that “the educational
experience and the culture of the discipline (as reflected in the attitudes and practices of [STEM]
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
557
faculty) make a much greater contribution to [STEM] attrition than the individual inadequacies of
students or the appeal of other majors” (p. 392). We extend this work by increasing the number and
variety of institutions that students attend, using quantitative methods, and by focusing on
diversity among STEM aspirants and their achievement of major goals from college entry.
Guiding our work are a recent set of studies that have helped to clarify the impact of unique
experiences and circumstances on STEM degree completion for underrepresented racial minority
(URM) students, as discussed below.
Key Individual Experiences and Institutional Factors
For URM students who pursue undergraduate studies in STEM fields, a combination of
environmental and internal psychological factors has been shown to impact persistence. For
example, Elliott et al. (1996) found that African American students had far less preparation in precollege sciences, as demonstrated by lower rates of participating in AP Biology, Chemistry,
Physics, and Calculus courses. As Russell and Atwater (2005) noted, a demonstrated competence
in science and mathematics at the pre-college level is vital to African American students’
successful progress through the science pipeline from high school to college. Other critical factors
include receiving family support and teacher encouragement, developing intrinsic motivation,
and maintaining perseverance. Likewise, the presence of family support and guidance from
faculty mentors are associated with the development of greater academic self-efficacy and success
in the sciences for Latino students (Anaya & Cole, 2001; Cole & Espinoza, 2008; Torres &
Solberg, 2001).
Beyond individual factors, campus environments play an important role in improving
undergraduate success in STEM fields. At the programmatic level, offering undergraduates
research opportunities can make a difference not only in attracting and retaining STEM majors,
but also in facilitating students’ learning in the classroom by introducing them to the benefits of
what science research careers might entail (Kinkead, 2003; Lopatto, 2003). URM students who
participate in well-structured undergraduate research programs receive added benefits as well,
including enhancing their knowledge and comprehension of science (Sabatini, 1997); clarifying
graduate school or career plans in the sciences (Hurtado, Cabrera, Lin, Arellano, &
Espinosa, 2009; Kardash, 2000; Sabatini, 1997); and obtaining other professional opportunities
that develop their scientific self-efficacy (Gándara & Maxwell-Jolly, 1999; Hurtado et al., 2009;
Mabrouk & Peters, 2000). According to Carlone and Johnson (2007), URM students who feel,
think, behave, and are recognized by meaningful others (e.g., faculty role models) as “science
people” are more confident about their academic abilities. Students of color who are more likely to
consider science as an important aspect of their self-identity, are also more likely to persist in their
STEM major (Chang et al., 2011; Espinosa, 2011).
Several recent studies further illuminate the individual experiences and institutional attributes
that can significantly improve URM students’ chances of completing an undergraduate degree in a
STEM field. In one longitudinal study of biomedical and behavioral science aspirants, Hurtado
et al. (2007) examined the impact of college on two key outcomesin the first year of college, sense
of belonging and academic adjustment. They found that the relevance of coursework to students’
lives had a positive impact on academic and social adjustment for White and Asian students as
well as for URMs in the sciences. Although this finding underscores the importance of experiential
learning and understanding the application of knowledge for all aspiring scientists, it may also
confirm previous studies, which found that URM students often leave the sciences due to a
perceived lack of relevance with their social values and desire to improve conditions for their
communities (Bonous-Hammarth, 2000). Moreover, campus climate may be at play as well.
Hurtado et al. found that perceptions of hostile racial climates were negatively associated with the
Journal of Research in Science Teaching
558
CHANG ET AL.
sense of belonging of all students, whereas such climates hindered the academic adjustment of
only URMs. Further, Hurtado et al. found that aspiring scientists of all racial groups were more
affected by financial concerns than their non-science counterparts, but URM science students
were most strongly affected by such concerns, which further inhibited both their academic and
social adjustment. Other national research studies have also identified effects of the campus racial
climate on retention in college (Museus, Nichols, & Lambert, 2008; Titus, 2006).
Similarly, in examining key factors that influence career aspirations in science research for
students entering their first year of college, Oseguera, Hurtado, Denson, Cerna, and Sáenz (2006)
found that entering URM undergraduates reported working more hours during high school
and were more likely to expect to work full-time during college than their White and Asian
counterparts. These authors argued that having financial concerns and misperceptions about
the financial viability of research professions can deter students from choosing a career in
STEM fields. In contrast, they also found that participating in research-oriented programs prior to
college substantially increased entering URM freshmen’s interests in pursuing a scientific
research career. Reinforcing these findings, Adedokun, Bessenbacher, Parker, Kirkham, and
Burgess (2013) found that research skills and research self-efficacy predict student aspirations for
research careers.
Unfortunately, opportunities for participating in research seem to be harder to come by for
URM students when in college. In another study drawing from a similar longitudinal data set,
Hurtado et al. (2008) found that African American students have significantly lower odds of
participating in health science research during college compared to their White counterparts.
However, African American students that attended institutions offering formal health science
research opportunities to first-year students were much four times more likely to participate in
research than students at institutions without such programs. Similar to findings from other
studies reported earlier, Hurtado et al. (2008) found that students’ concerns about financing
college had a significant influence on research participation. African American students who
indicated having more serious financial concerns about paying for college were significantly less
likely to participate in health science research than their peers who were less concerned about
finances.
Another set of studies examined factors that contributed to persisting in a STEM major
through a student’s first-year of undergraduate study. Chang et al. (2008) found that aspiring to
attain a graduate degree increased URM students’ likelihood of staying in a biomedical science
major through the first-year of college by over 30%. More impressively, joining a pre-professional
or departmental club during a student’s freshman year increased the likelihood of persisting by
more than 150%. This study also found that a URM student had a 30% higher chance of departing
from a science major if he or she attended an institution where the average undergraduate
combined math and verbal SAT score (the “selectivity” of the institution) was 1,100, versus one
with an average of 1,000. In addition, they also found that a 100-point average undergraduate SAT
score increase at the institution-level lowered the chances of biomedical persistence by 20% for all
students. Similarly, Espinosa (2011) found that after 4 years of college, for every 100 point
increase in institutional selectivity, women of color in STEM were almost half as likely as White
women to persist in a STEM major (7.6% compared with 14.0%). So, higher institutional
selectivity (measured by student achievement scores) negatively affects all science students, but
the effect is stronger for URM students. Curiously, this effect does not appear to apply to those
students who attended historically Black colleges or universities (HBCUs) in the first year of
college; for these students the opposite tended to occur. That is, as the average undergraduate
combined SAT score increased, the chances of persisting in STEM for students attending HBCUs
also tended to improve (Chang et al., 2008). These studies indicate the moderating effects of
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
559
selective contexts associated with race and STEM persistence in the first year of college, and
with gender in the fourth year of college. The current study extends this work by examining
contextual selectivity effects for all STEM aspirants, especially URM students, to the fourth year
of college.
One shortcoming concerning some of the studies reviewed here is that they employed singlelevel statistical techniques that did not account for the multi-level nature of the data (e.g., Cole &
Espinoza, 2008; Elliott et al., 1996; Grandy, 1998). Data on the college student experience is by
nature multi-level, as students are “nested” within institutions. Analytical techniques that do not
account for this nesting are not only less robust but also risk drawing erroneous conclusions due to
misestimated standard errors. The present study further adds to the existing literature on
undergraduate STEM persistence by utilizing a more robust analytical technique that can account
for the multi-level nature of student data. By accounting for the institutional characteristics and
student level characteristics separately, this study can more accurately assess the important
contextual factors contributing to STEM degree persistence.
Taken together, the findings reported above are nicely captured in Nora et al.’s (2005) model
of student persistence and degree attainment. The Nora et al. model is a reformulation of the Tinto
model (1993) that brings more clarity to the academic dimensions of the college environment
while maintaining social and academic integration as a central tenet. Based on research on
underrepresented groups of students, Nora et al.’s integration model includes many of the factors
that are likely to influence minority, low-income, and non-traditional student populations in
important ways—such as aspects of pre-college socialization environments (school and home
environment), financial assistance/need, family support, environmental pull factors (family
and work responsibilities), and commuting to college. In reference to the academic and social
experiences in college, the model emphasizes such experiences as formal and informal academic
interactions with faculty, campus climates (perceptions), validating experiences (from faculty and
peers), and mentoring relationships (faculty, peer, and advising staff). The model also emphasizes
academic performance, academic/intellectual development, and non-cognitive gains (in psychosocial domains) as intermediate outcomes, which in turn determine persistence in college. In the
current study, we incorporated measures from many of these key areas to help explain persistence
in STEM at the college students entered as first-time freshman.
Hurtado (2007) suggests that sociological models of college impact should include four
measurable domains of institutional and normative constructs: characterizations of the environment focusing on student perceptions of their experiences within the social and academic systems
of the collegiate environment; social interactions that capture both the frequency and quality of
informal academic and social engagement in college; formal memberships based on both
individual interest and how the group determines entry and confers privileges on its members; and
perceived social cohesion or the students’ own psychological sense of integration in the college
community. In multi-institutional studies, it is thus important to include relevant structural
characteristics that define distinctions between colleges, such as minority enrollment and
selectivity, which may further account for probabilities of STEM retention in particular types of
institutions.
Building on studies of STEM student persistence and transitions in the first year (Chang et al.,
2008, 2011; Hurtado et al., 2007), we adopted key constructs based on the Nora et al. (2005)
integration model to detail the link between persistence in STEM to the fourth-year of college and
student experiences at multiple types of 4-year colleges. Specifically, we tested the hypothesis that
STEM persistence is not only a function of the characteristics students bring at college entry, but is
also affected by participation in formal structures that distinguish undergraduate experiences, the
racial dynamics of a college, the continuing influence of family, financial concerns, and student
Journal of Research in Science Teaching
560
CHANG ET AL.
assessments of their own development and competence in their identity as a scientist. We apply
this understanding to examine the extent to which racial disparities in science achievement can be
explained by those and other factors that can then be applied to improve URM STEM degree
persistence.
Methods
Research Questions
This study was designed to examine individual and institutional factors that predict 4-year
STEM degree persistence, with a focus on underrepresented racial minority students. Two main
lines of questioning guided this study. Specifically, we asked:
1. Among all students who started college with an interest in majoring in a STEM field,
does a student’s race contribute significantly to the chances that he or she will follow
through on these intentions? If so, are the effects of race moderated by high school
academic preparation and/or key college experiences?
2. If there are racial disparities in persistence rates after controlling for pre-college student
characteristics, what are the college factors that contribute to the persistence of URM
students? That is, what college experiences and institutional characteristics significantly
predict the likelihood that a URM student will follow through on his or her freshman year
intentions to pursue a degree in STEM?
Data and Sample
Data for this study were drawn from the Cooperative Institutional Research Program (CIRP)’s
2004 The Freshman Survey (TFS) and 2007–2008 College Senior Survey (CSS). The CIRP is a
program of data collection and research housed at the Higher Education Research Institute at the
University of California, Los Angeles. The TFS and CSS are administered annually by CIRP to
college students across the U.S., and each survey collects a wide variety of information about
students (see Liu, Ruiz, DeAngelo, & Pryor, 2009 and Sax et al., 2004 for more information about
these surveys). The 2004 TFS was administered to first-year students entering college in the
summer/fall of 2004, either during freshman orientation or during the first few weeks of the fall
term. The 2008 CSS followed up with this same group of students in the spring of or summer after
their fourth year in college. The 2008 CSS data were linked to the 2004 TFS data to form a
longitudinal dataset that tracked students over their first 4 years of college. The overall
longitudinal response rate for the TFS-CSS was 23%. Response weights were applied to the data
to reduce nonresponse bias, as is detailed in the analyses section. To the longitudinal database, we
added institution-level data from the Integrated Postsecondary Education Data System (IPEDS),
drawn from academic year 2006 to 2007.
Grants from the National Institutes of Health (NIH) and National Science Foundation (NSF)
provided funds for a targeted sampling strategy for this study. In addition to the large number of
institutions that typically participate in CIRP surveys, we supplemented the TFS sample using an
NIH grant that allowed for the specific recruitment of students at minority-serving institutions that
have strong reputations of graduating undergraduates in the biomedical and behavioral sciences.
The funding also allowed us to target students at institutions that have NIH-funded undergraduate
research programs. Additional funding from NSF allowed us to expand the regular CSS sample to
include students at institutions that have strong reputations for producing bachelor’s degrees in
STEM. The overall goal of our data collection strategy was to obtain a large and diverse sample of
students from underrepresented racial and ethnic groups who were interested in STEM, as well as
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
561
a set of their White and Asian counterparts for comparison, in order to assess how science
achievement is differentiated by race across different types of institutions.
The current study used two samples of students to answer its main research questions. For the
first research question, which asked whether a student’s race significantly predicts the likelihood
of persisting in a STEM degree, we used all students who took both the TFS and CSS (at the same
institution), and who indicated on the TFS that they intended to major in a STEM field. This
sample included 3,670 students at 217 different institutions. One thousand five hundred twentytwo of the respondents (41.5%) identified as White (58.0% of whom were female), 498 (13.6%) as
Asian (59.9% female), 812 (22.1%) as Latino/a (60.0% female), 626 (17.1%) as Black/African
American (72.9% female), and 196 (5.3%) as Native American (55.9% female). Overall, the
sample was approximately 61.3% female. For our second research question, we restricted the
above sample to include only underrepresented racial minority (URM) students—that is, only
students who indicated they were Native American, Latino/a or Black/African American. In total,
1,634 students were included in the URM subsample; 64.6% of these students were female.
Appendix A shows descriptive statistics for URM students.
Variables
Nora et al.’s (2005) theoretical model and the research reviewed earlier informed the selection
of the independent variables in the model (see Appendix A). The dependent variable used in this
study was dichotomous and represented whether students who graduated or were still enrolled
after 4 years of college had followed through with their first-year intentions to pursue a degree in a
STEM field (coded 1), or whether they switched majors and completed or continued to pursue a
degree in a non-STEM field (coded 0). The selected variables included student demographics and
background characteristics, college experiences, and institutional characteristics. Student-level
characteristics were grouped into several blocks to aid analysis and interpretation. Specifically, we
grouped student background characteristics into (a) demographics (race, sex and socioeconomic
status [as proxied by mother’s education level]); (b) high school academic preparation (grades,
SAT score, high school course taking patterns), and (c) other pre-college characteristics (including
degree aspirations, concern about financing higher education, and student assessments of their
academic and social strengths). Finally, all college experiences were included as one block, and
within this block there were measures corresponding to students’ interaction with faculty,
assessments of the institutional climate, psychosocial concerns, and social and academic
integration and involvement.
In addition to the student-level variables, institution-level variables were also modeled. These
included institutional type and control (4-year/university, public/private), institutional selectivity
(measured by the average math and verbal SAT score of entering freshmen), percent of students
majoring in STEM fields, structural diversity (percent of student body that is Black, Native
American or Latino/a), whether an institution is a historically Black college or university
(HBCU), proportion of students receiving financial aid and/or federal aid, and institutional size (as
measured by total undergraduate FTE).
Analyses
Missing Data. Before conducting our main analyses, we first addressed missing data. We
began by using listwise deletion to remove all cases for which no information was available on
the outcome variable, demographic characteristics, and/or dichotomous college experiences
(i.e., participation in undergraduate research programs or clubs relating to a major, working full
time while in school). For the remaining variables in the model, we analyzed the extent to which
Journal of Research in Science Teaching
562
CHANG ET AL.
missing data occurred. Overall, there was very little missing data. No variable had more than 6%
of cases missing, and examination of missing data patterns suggested that missing data occurred at
random. The SAT variable had the highest proportion of missing data, at 5.1%. Most variables
were missing data in fewer than 1% of cases.
Given the relatively few instances of missing data across the variables used in the analysis, we
elected to fill in missing data using the expectation maximization (EM) algorithm. The EM
algorithm employs maximum likelihood estimation techniques to impute values for cases with
missing data, and because it uses most of the information available in the dataset to produce
the imputed values, it is a more robust method of dealing with missing data than listwise
deletion or mean replacement (Allison, 2002; Dempster, Laird, & Rubin, 1977; McLachlan &
Krishnan, 1997).
Weighting. Because the longitudinal response rate for the TFS-CSS sample was only 23%, we
calculated and applied response weights to the data to adjust for any non-response bias that might
be present.1 The aim of this weighting was to adjust our CSS sample of respondents to resemble
the original population targeted by the CSS—that is, the TFS participants (Babbie, 2001). All
analyses performed for this study were conducted using weighted data.
Multivariate Analyses. The clustered, multi-level nature of our data and the dichotomous
outcome variable warranted the use of hierarchical generalized linear modeling (HGLM). HGLM
is an ideal statistical technique for this study, as it can separate individual and institutional effects
to help determine how individual characteristics interact with institutional contexts to affect
STEM major persistence. Further, performing single-level analyses with multi-level data can
underestimate the standard errors of model parameters, which can inflate Type-I statistical error
(de Leeuw & Meijer, 2008; Raudenbush & Bryk, 2002). To ensure the use of HGLM was justified,
we first ran a fully unconditional model, or a model with no predictors specified at level one or
level two. This unconditional model allowed us to assess whether there was significant variation in
the outcome between institutions, via an examination of the variance of the random effect at level
two. That is, the significance of the random effect variance at level two of the unconditional model
(the between-institution variance component) allowed us to assess whether students’ average
probabilities of persisting in STEM majors significantly varied across institutions; if they did,
the use of HLM would be justified. For both the whole sample and for the URM subsample, we
confirmed that these institutional probabilities of persistence varied significantly, so we proceeded
with our modeling course. All of our HGLM analyses were performed using HLM 6.0
(Raudenbush, Bryk, Cheong, & Congdon, 2004).
When using hierarchical modeling such as HGLM, researchers must make choices regarding
the centering of variables. Because we were interested in the average effect of each predictor on
students’ likelihood of persisting in STEM, we chose to grand-mean center all continuous
variables, as well as all ordinal categorical variables with more than two categories. Grand-mean
centering subtracts the mean of the variable for the entire sample from each individual observation,
and allows the model intercept to be more easily interpreted (Raudenbush & Bryk, 2002).
Dichotomous variables were left un-centered. In order to easily interpret the results of the final
model for URMs, we reported the HGLM results for significant predictors as delta-p statistics,
calculated using the formula provided by Petersen (1985). These statistics represent the expected
change in probability of persisting in a STEM major (vs. not persisting) that is associated with a
one-unit change in the predictor variable. For dichotomous predictor variables, we calculated
delta-p using the mean of the sample as an estimate of the probability of STEM persistence for the
reference group. For all other predictors, we used the sample mean to estimate the probability of
persisting when evaluated at the mean of the predictor variable. For these non-dichotomous
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
563
predictors, the delta-p statistic represents the expected change in probability associated with oneunit change from the mean (Pampel, 2000).2
Full Sample HGLM Models. To address research question 1 we ran three models predicting
STEM persistence using the full sample of undergraduates who indicated on the TFS that they
intended to major in a STEM field. Although these models did not contain any level two predictors,
they were run in HLM 6.0 to ensure that the clustering of students within institutions was taken
into account, thereby avoiding misestimation of standard errors. In our first model, we utilized
only race/ethnicity, gender, and mother’s level of education (a proxy of socioeconomic status) to
predict STEM persistence. In the second, we added variables to the model representing high
school academic preparation—grades, SAT score, and indicators of years of STEM course-taking.
Finally, we removed the high school preparation variables and added a set of variables representing
experiences in college that may contribute to (or detract from) STEM persistence. The general
form for each of the full-sample HGLM models was as follows:
Level 1 :
wij
log
1 wij
!
¼ b0j þ b1j X 1ij þ b2j X 2ij þ þ bQj X Qij
Level 2 :
bQj ¼ g q0
ð1Þ
ð2Þ
where wij is the probability of persisting in STEM, there are i ¼ 1, . . ., I level-1 units (students)
nested within j ¼ 1, . . ., I level-2 units (institutions), and there are Q predictors at level 1 and no
predictors at level 2—that is, the level-1 effects are constant across level-2 units. Notationally,
there are Q þ 1 coefficients at level 1, and:
b0j is the intercept at level 1,
gq0 is the intercept at level 2;
bQj are level-1 coefficients for
XQij level-1 predictor variables; and
gq0 are level-2 fixed effects.
Appendix C lists each predictor that was included in the three models run on the full sample.
URM-Only HGLM Models. Our second research question concerned the impact of college on
URM STEM persistence rates. The model building for this analysis proceeded in several stages,
mirroring the full-sample analysis. Specifically, we first modeled STEM persistence with
only student background characteristics as predictors. We then added college experiences, and
finally institution-level characteristics. For the analyses without institution-level predictors, the
general form of the HGLM models were the same as given above in Equations (1 & 2). For the final
model, we allowed the intercept of level 1 (b0j) to vary, in order to examine how institutional
contexts impact STEM persistence. The general form for the level-1 equation in our final model
was the same as shown in Equation (1). The general form of the level-2 equation for the intercept
b0j was:
b0j ¼ g 00 þ g 01 W 1j þ g 02 W 2j þ þ g 0S q W S q j þ uqj
ð3Þ
Journal of Research in Science Teaching
564
CHANG ET AL.
where there are Sq predictors ðW S q j Þ and Sq þ 1 coefficients ðg 0S q Þ at level 2, uqj is the normally
distributed level-2 error, and for all other level-2 predictors,
bQj ¼ g q0 :
ð4Þ
The specific predictors used in the final URM-only analysis are detailed in our results tables
and in Appendix A.
Limitations
Before presenting and interpreting the results of our analyses, it is important to note some of
this study’s limitations. First and foremost, our sample included only students who were still
enrolled or were graduating after 4 years of college. In other words, students who withdrew or
stopped out were not included in the sample, and thus our results apply only to those students who
were successful in persisting in the same college they entered as freshmen–though not necessarily
in their original intended major. In addition, our study had a relatively low longitudinal response
rate (23%), and the extent to which our results are generalizable to a larger group of students may
thus be limited. Although we attempted to correct for the nonresponse bias that may have been
introduced by the low response rate, our correction was limited to the entering characteristics we
had available on the TFS, and may not have taken all relevant factors into consideration. Further,
our sample contained relatively few Native American students—just under 200—so our analyses
had limited statistical power to detect significant results for this group.
Another limitation of our study is that we define “STEM persistence” as “following through
on first-year intentions to major in STEM.” Entering freshmen who take the TFS may not have a
good sense of all of the areas of study available to them, and thus some students who initially
thought they would major in STEM may fail to follow through on these aspirations because they
discovered another field that better aligned with their evolving interests. Still, because our findings
(discussed below) mirror previous results, especially regarding differences in persistence rates by
race, it appears that our calculation of STEM persistence is capturing an important aspect of
movement in the STEM pipeline. More importantly, no other national data set contains measures
of students’ aspirations upon college entry, so we are uniquely positioned to understand loss of
science talent prior to and after students select their major.
Results
Descriptive Statistics: STEM Persistence
Among the aspiring scientists in our overall sample, 62.5% persisted in a STEM major to the
fourth year of college. This rate was lower among URM students (58.4%) than it was among Asian
American (73.5%) and White (63.5%) students. Disaggregating by URM groups, African
Americans had the lowest rate of STEM major persistence (56.5%), followed by Latino/as
(58.9%) and Native Americans (62.8%).
Full Sample HGLM Results
To answer research question 1, three separate hierarchical models were conducted using the
full sample of undergraduates who indicated on the TFS that they intended to major in a STEM
field (see summary of results in Table 1 and detailed results in Appendix B). Model 1 analyzed the
effects of racial classification (i.e., self-identification as Native American, Latino, or African
American) on 4-year STEM persistence, controlling for two key demographic characteristics
(gender and mother’s level of education). We found significant effects for both African Americans
and Latinos; students who identified as being from either of these groups were significantly less
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
565
Table 1
Hierarchical generalized linear modeling (HGLM) results for all students, looking at main effects of race
All Students (N ¼ 3,670)
Model Number:
1
Race main effects (Whites and Asians are reference group)
Native American
Latino
Black/African American
Blocks of variables included in the model
Other demographic characteristics (gender, mother’s education)
High school academic preparation
College experiences
2
3
No
Yes
Yes
Significant effects?
No
No
No
No
No
No
X
X
X
X
X
Effect significant, p < 0.05, see Appendix C for detailed statistics and lists of variables in each block.
likely to persist in STEM majors than were their White and Asian American counterparts. To
examine whether these racial differences in STEM persistence were due to differential high school
academic achievement and/or experiences, we next added a set of variables to the model
representing high school academic preparation (grades, SAT score, and high school course taking
patterns, Model 2). When the high school preparation and experience variables were included as
predictors along with the demographic characteristics, the race variables were no longer
statistically significant, suggesting that high school preparation moderates the effects of race.
We next examined whether experiences in college might also moderate the negative race
effects observed in Model 1, by removing the high school preparation variables from the model
and adding in college experiences (Model 3). Encouragingly, we found that college experiences
also moderated the effects of race—after taking college experiences into account, the race
variables did not exhibit a significant effect on STEM persistence. This indicates that colleges and
universities can play a significant role in moderating racial disparities in STEM persistence.
To more accurately identify what key experiences might be for URM undergraduates, we
conducted another set of analyses with the sample of only URM students. As described above, the
model building for this analysis proceeded in several stages, with student background characteristics entered first, followed by college experiences, and finally by institution-level characteristics.
As the results changed relatively little from one stage to the next, the HGLM model for URM
students is presented only in its final form.
URM Subsample HGLM Results
Table 2 presents the results from the URM-only HGLM analysis examining the impact of
background and college experiences on the likelihood of persisting in STEM majors. Focusing
first on student demographic characteristics, we found no significant difference in the probability
of STEM persistence when comparing Latinos and Native Americans to African Americans.
Student gender and SES (proxied by mother’s education) also had no significant effects. Among
the high school academic preparation predictors, only one variable was significantly associated
with STEM persistence for URM students: SAT score. For a 100-point increase over the mean in
combined math and verbal SAT score, our model suggests that URM students will be 6.86
percentage points more likely to persist in a STEM degree. Curiously, high school GPA and the
number of years students took mathematics, physical science, and/or biological science in high
school did not significantly affect undergraduate STEM persistence over and above other
predictors in the model.
Journal of Research in Science Teaching
566
CHANG ET AL.
Table 2
Hierarchical generalized linear modeling (HGLM) results for URM persistence in a science, technology,
engineering or math (STEM) major
URM Students Only (N ¼ 1,634)
Demographic characteristics
Native Americana
Latinoa
Student’s gender
Mother’s Education
High school academic preparation
Average high school grade
Math þ Verbal SAT score (in 100-point increments)
Mathematics
Physical science
Biological science
Participated in Summer Research Program
Other pre-college characteristics
Intend to obtain Master’s degree (M.A., M.S., etc.)b
Intend to obtain Ph.D. or Ed.D.b
Intend to obtain M.D., D.O., D.D.S., D.V.M.b
Entering science identity
TFS academic self-concept
TFS social self-concept
Concern about financing college education
College experiences
Worked full-time while attending school
Felt family support to succeed
Faculty mentoring factor
Asked a professor for advice outside of class
Felt intimidated by your professors
Faculty feel that most students here are
well-prepared academically
Studied with other students
Sense of belonging
Positive cross racial interaction
Negative cross racial interaction
Hostile racial climate
There is strong competition among most of the
students for high grades
Felt overwhelmed by all I had to do
Participated in an undergraduate research program
(e.g., MARC, etc.)
Joined a club or organization related to your major
Worked on a professor’s research project
Institutional characteristics
Institutional control (Private keyed higher)
Institutional type (4-year keyed higher)
HBCU (HBCU’s keyed higher)
Selectivity in 100-point increments
Percent of students on financial aid
(in 10-point increments)
Log Odds
SE
0.16
0.00
0.08
0.04
0.25
0.16
0.16
0.04
0.08
0.29
0.08
0.03
0.02
0.16
0.06
0.07
0.14
0.06
0.08
0.19
0.06
0.01
0.46
0.06
0.04
0.03
0.14
0.21
0.27
0.23
0.10
0.01
0.01
0.12
0.39
0.11
0.27
0.08
0.13
0.08
Dp
Sig.
6.76
0.00
11.50
0.04
0.98
0.81
0.00
0.00
0.18
0.13
0.09
0.12
0.13
0.12
9.74
0.03
6.79
0.00
0.60
0.17
0.12
0.21
0.01
0.14
0.12
0.09
0.09
0.11
0.10
0.09
13.57
0.00
0.09
0.80
0.15
0.22
17.38
0.00
0.40
0.19
0.14
0.12
9.32
0.01
0.39
0.07
0.20
0.52
0.03
0.32
0.27
0.76
0.12
0.06
13.00
0.00
(Continued)
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
567
Table 2
(Continued)
URM Students Only (N ¼ 1,634)
Percent of students receiving federal aid
(in 10-point increments)
Indicator of whether institution has any research
expenditures (>$0)
Percent of students majoring in STEM
(in 10-point increments)
Percent of student body that is American Indian,
Black, or Latino
Log(Undergraduate FTE)
Intercept
Model statistics
Chi-square
Intercept reliability
Explained variance at level 2
Baseline probability of STEM major persistence
Log Odds
0.03
SE
0.09
0.01
0.26
0.23
0.06
0.06
0.10
0.10
0.87
0.18
1.08
Dp
5.57
Sig.
0.00
0.09
0.694
0.58
a
Reference group for race variables is Black/African American.
Reference group for degree aspiration variables is Bachelor’s degree only.
b
Three more pre-college variables were significantly associated with the likelihood of
persisting in a STEM major. The largest of these concerned the intent to obtain the following
profession/clinical science-oriented degrees: M.D., D.O., D.D.S., and D.V.M. URM students who
came to college with intentions of pursuing one of these professional degrees were 11.5 percentage
points less likely to persist in a STEM field than were those who intended to obtain only a
Bachelor’s degree. URM students who came in with aspirations for a Master’s degree or Ph.D./
Ed.D., on the other hand, were no more or less likely than their Bachelor’s-aspiring counterparts to
persist in STEM majors. Entering students’ academic and social self-concepts also significantly
predicted the likelihood of STEM persistence. Having a higher academic self-concept when
beginning college positively predicted persistence, whereas having a higher social self-concept
negatively predicted persistence.
Five college experiences significantly predicted the likelihood of URM students following
through on their freshman intentions to major in STEM. The strongest of these predictors was
participation in an undergraduate research program. URM students who participated in programs
that exposed them to research were 17.4 percentage points more likely to persist in STEM than
those who did not. Similarly, a positive though less striking effect was also shown for joining a
club or organization related to students’ majors. Students who joined such organizations were 9.3
percentage points more likely to persist in STEM than those who did not. Likewise, URM students
who studied more frequently with other students were also more likely to persist in STEM; we
observed a 13.6 percentage point improvement in chances of persisting for a one-unit increase
from the mean.
Two college experiences exhibited negative relationships with persistence. Students who
worked full-time while attending school (at any point in their college career) were 9.7 percentage
points less likely to follow through with intentions to major in STEM, compared to those who
never worked full-time. Second, students who had more mentoring from faculty (as measured by
our faculty mentoring factor, see Appendix B) were more likely to have chosen a major outside of
STEM than were students who reported lower levels of mentoring activity with faculty. Because
Journal of Research in Science Teaching
568
CHANG ET AL.
this finding seems counterintuitive, we examined it further and found that this factor had no impact
until we controlled for student participation in a faculty member’s research project. In other words,
the initial relationship between the outcome and the faculty mentoring variable was nonsignificant, suggesting that students who persisted in STEM were just as likely as those who did
not persist to participate in a range of mentoring activities with faculty. However, after controlling
for participation in research, which is a major determinant for students persisting in STEM, we
found that students who persisted in STEM tended to report lower rates of faculty mentoring
activities than students who switched to non-STEM fields. Thus, our results seem to show that
faculty mentoring on its own is not necessarily associated with STEM persistence, but rather
that its relationship with persistence is contingent on whether or not the student participated in
research.
At the institution level, two measures of college context were found to be significant
predictors of STEM persistence. First, the proportion of the student body majoring in STEM fields
at an institution significantly and positively contributed to the average likelihood of STEM
persistence at that institution. For a 10-point increase from the mean proportion of undergraduates
majoring in STEM, the average likelihood of a URM student persisting in STEM was predicted to
increase by 5.6 percentage points. However, on the negative side, institutional selectivity, as
measured by the average math plus verbal SAT scores of entering students, negatively predicted
STEM persistence. For a 100-point increase from the mean institutional selectivity, the average
likelihood of STEM major persistence dropped by 13.0 percentage points. We found no significant
effects of institutional control, type, HBCU status, research expenditures, structural diversity, or
size. The institutional predictors accounted for 69.4% of the between-institution variance in
students’ average probability of following through with their initial intentions to major in STEM.
Discussion
Although the racial disparities in science achievement have been well documented, it is not
altogether clear why this problem persists and what institutions of higher education can do about it
(Mutegi, 2013). The findings from this study begin to untangle some of those issues and suggest
that the effect of race on persistence in a STEM major is largely associated with unequal
preparation and access to educational opportunities. While our findings show that identifying as
African American or as Latino, compared to identifying as Asian American or White, was
associated with a lower likelihood of STEM persistence, the negative effects of race were
moderated by both pre-college characteristics and college experiences. In other words, addressing
key factors either before students enter college or while they are in college can independently and
significantly diminish the racial disparities in STEM achievement.
Regarding pre-college characteristics, we found that across a wide range of institutions,
having higher SAT scores and a higher academic self-concept as an entering freshman contributed
in positive ways to the persistence of underrepresented racial minority (URM) undergraduates
(Latinos, Native Americans, & African Americans) in STEM fields. These findings are consistent
with those of previous studies, which have pointed to the importance of students’ high school
academic preparation (Elliott et al., 1996; Russell & Atwater, 2005). Thus, our findings support
ongoing calls for early intervention that build and sustain science self-efficacy. Those
interventions have more positive long-term effects when they are reinforced with activities in
college that ensure student success (e.g., bridge programs, supplemental instruction, and tutoring,
and meta-cognitive study strategies) and assist students through introductory coursework. Further,
enhanced support for performing well on tests in terms of curriculum content and test preparation
are also likely to reduce the racial disparity in science achievement. Unfortunately, institutions of
higher education are arguably less well-positioned to directly address the persistent racial
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
569
disparities in high school preparation, so the discussion below emphasizes the findings from this
study that can be applied more directly by colleges and universities. This focus offers a distinctive
contribution to the developing literature and practices on STEM retention, particularly for
underrepresented groups. The discussion will conclude with a set of recommendations tied to our
findings.
One unique finding from our study regarding pre-college characteristics is that those entering
URM freshmen who aspire to obtain a professional/clinical science-oriented degree (M.D., D.O.,
D.D.S., or D.V.M) were less likely to persist in STEM than were their counterparts who did not
report the same degree intentions. This finding is especially noteworthy because professional/
clinical degree aspirations are popular among URM students who chose STEM fields as freshmen
(almost 30% of our URM STEM aspirant sample reported such degree aspirations at college
entry). One explanation for this finding could be that interest in pursuing professional/clinical
degrees, especially medicine, tend to be more competitive because those professions are widely
viewed as socially prestigious and financially rewarding; they may thus attract and screen out
more talent. Another possible explanation could be that entering freshmen interested in obtaining
those types of degrees have only a superficial understanding of those related professions and that
such aspirations signal a weak interest and commitment to science learning. It will be important to
more closely monitor this group of students, as we point out in the recommendations that follow.
A number of our findings show that exposure to unique college experiences can make a
significant positive difference in STEM degree attainment for URM undergraduates. Consistent
with other studies, we found that URM students who participated in an undergraduate research
program increased their chances of obtaining or continuing to progress toward completing a
STEM degree, by an impressive 17.4 percentage points. According to Carlone and Johnson’s
(2007) qualitative study of women of color, students in research programs are typically given the
opportunity to engage in the practical application of their coursework, which improves their
degree of identification with science through “performance and competence.” As students feel
more personally connected to STEM, Carlone and Johnson argue, they are more likely to persist in
their respective majors. Students in research programs are also connected to committed faculty
who nurture students in ways that enable them to take on progressively complex research tasks and
subsequently identify as scientists (Hurtado et al., 2009).
We also found that those URM students who joined a club or organization related to their
major significantly improved their chances of persisting. Clubs such as the National Society of
Black Engineers (NSBE) and the Society of Women Engineers (SWE) are examples of
undergraduate student organizations that revolve around particular STEM majors. These clubs
provide their targeted membership with a wide range of academically enriching experiences,
which promote socially and academically supportive networks. Perhaps by providing more
opportunities to engage meaningfully with other students who share similar academic interests
and trajectories, institutions also provide more opportunities for students to study together.
Applying peer learning strategies, for example, have been found by Lopez, Nandagopal,
Shavelson, Szu, and Penn (2013) to further improve academic performance in STEM courses.
Overall, these findings confirm previous calls (Espinosa, 2011) to build a stronger social network
of peers in STEM majors.
Further attention to student finances will also be key in organizing interventions for STEM
degree persistence. We found that working full-time while attending school is negatively
associated with URM students’ chances of persisting in a STEM field. Other studies (DeAngelo,
Franke, Hurtado, Pryor, & Tran, 2011; Titus, 2006) have indicated that having more financial
concerns and fewer resources during college present obstacles for degree completion, but our
finding uniquely show that this is also the case for URM STEM aspirants. While working full-time
Journal of Research in Science Teaching
570
CHANG ET AL.
may relieve some financial burdens, it may also put greater strain on students’ time to invest in
their coursework and science career aspiration. It thus seems that STEM persistence interventions
should be structured to relieve students of the burden of working full-time—for example, by
providing stipends for work on research projects.
Curiously, our model showed that higher-levels of faculty mentorship were associated with a
higher probability of switching to a non-STEM major. Upon further examination, this finding was
explained by the fact that participating in faculty research was strongly associated with receiving
mentorship; once research experiences were taken into account in our model, lower faculty
mentoring was evident among URM STEM persisters. This suggests that faculty mentoring in its
current form may be limited at many institutions to specific programs and research experiences.
Perhaps for URM students who do not participate in research or special programs, the reason to
seek mentoring from faculty might be related to difficulties with coursework; students having
difficulties may be more likely to switch out of a STEM major. In any case, institutions will need to
more carefully consider faculty’s role—beyond providing research opportunities—in improving
URM students’ chances of achieving their intended academic goals, especially since
faculty recognition is an important component of STEM identity development (Carlone &
Johnson, 2007).
Lastly, we tested and advanced Seymour and Hewitt’s (1997) claim that institutional context
makes a difference in URM students’ chances of following through with their intent to degree in a
STEM field. As Seymour and Hewitt observed in their qualitative study, this effect may be related
to the educational experiences and the culture of STEM disciplines, which are especially unique
compared to other areas of study. Indeed, we confirmed the importance of institutional context
using a much larger sample of STEM aspirants than previous studies, which enabled us to control
for students’ predispositions while examining variations across different contexts. We found that
the percent of students at an institution who are majoring in a STEM field is positively associated
with URM students’ major persistence. Institutions with larger proportions of STEM majors
might have stronger STEM cultures that may provide fewer distractions to, and better support
students’ STEM related career goals, thereby contributing to the likelihood of a student
completing a STEM degree. We also found that the average student body SAT score, or college
selectivity, had a negative effect on persisting in a STEM field. A similar effect was documented
by Chang et al. (2008) on all first-year biomedical students including White and Asian students,
and by Espinosa (2011) on both White women and women of color after 4 years of college.3 Given
these findings concerning college selectivity, this issue may have more to do with the context of
science instruction and progress within science majors at selective institutions than with the
backgrounds of students. Although it is unclear how exactly selectivity or the proportion of
science majors affects persistence, the effects are likely a function of the peer culture in science,
institutional practices, and faculty roles as institutional agents that foster development. Our
findings thus point to the importance of creating an institutional culture of commitment to and
success in science.
Recommendations
While we confirmed that URM students are significantly less likely than their counterparts to
persist in a STEM major, our findings also suggest several key moves that institutions can adopt to
reduce racial disparities and increase the overall number of science degree recipients. Although
URM students’ pre-college characteristics, especially prior academic preparation, do make a
difference in their chances of following through with the intent to complete a degree in a STEM
field, we focus below on what colleges and universities can do once students begin their course of
study.
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
571
First, institutions can provide URM students with more extracurricular opportunities to
engage meaningfully in their chosen major. These opportunities can be in the form of wellstructured research programs or student organizations that enable students to be more engaged,
including major-related extracurricular activities, work on a professor’s research project, peer
study groups, or other social networks that support access to information and strategies for
navigating a STEM major. To have even more of an impact, extracurricular major-related
opportunities should also address financial concerns of students to obviate the need to work long
hours at jobs off campus. Since engagement in research requires a high level of commitment,
working too many hours for pay outside of those commitments can compromise students’ level of
engagement toward their degree aspiration. Both the National Institute of Health and the National
Science Foundation have supported programs that build in these opportunities, including, for
example, Minority Access to Research Careers (MARC), Minority Biomedical Research Support
(MBRS), and Research Experiences for Undergraduates (REUs). Opportunities to “earn and
learn” in STEM are ideal solutions for improving student success.
Second, faculty and departments should more closely monitor the success of those URM
students who aspire to obtain a professional/clinical science-oriented graduate degree (M.D.,
D.O., D.D.S., or D.V.M). Since such degree aspirations are popular among URM undergraduates,
yet are associated with lower rates of STEM degree completion, more effort should be made to
retain URM students in STEM majors even if aspirations to an M.D., D.O., D.D.S., or D.V.M. are
derailed. Perhaps institutions can better leverage professional/clinical degree aspirations to
“hook” students into the sciences early on in their studies, and can then broaden their
understanding about other careers for STEM graduates. Efforts related to providing more career
guidance in STEM fields might improve students’ willingness to continue working toward a
STEM degree rather than leave the sciences altogether when they abandon their aspiration to
pursue a professional/clinical degree.
Third, institutions need to examine more closely how their educational context contributes to
or obstructs from completing a STEM degree. Highly selective institutions, ironically, have higher
probabilities of degree completion but lower probabilities of STEM retention. By contrast,
institutions that enroll large proportions of students who are STEM majors are graduating a larger
proportion of those majors, and that success likely helps attract more students interested in
pursuing STEM degrees. By encouraging educators to look more closely at different educational
environments, and by having institutions engage in self-examination, we may be able to gain
greater insights into how to develop stronger and more inclusive “cultures of science.” The
complex dynamic of selectivity and STEM persistence should be of great concern because highly
selective colleges presumably enroll the most promising STEM students, and typically have more
resources to support student learning and advancement in science. One issue at selective
institutions that needs more attention is whether and how courses taken during the early years of
study (i.e., introductory chemistry, biology and calculus) promote STEM talent so that students
persist to the degree. As it stands, introductory courses at more selective institutions tend to serve
as “gatekeepers” that screen out more than nurture talent (Gasiewski, Eagan, Garcia, Hurtado, &
Chang, 2012), and the courses often encourage competition that discourages students from
continuing in the sciences.
Overall, our findings suggest that colleges and universities can make a significant difference
in reducing racial disparities in science achievement and do not have to wait idly for high schools
to send them more well-prepared students. Much more can be done at the undergraduate level to
shape students’ experiences and level of engagement in science, and to improve institutional
circumstances for students, especially for those underrepresented students who may be diverted
from their intended academic goals. The findings from this study identify promising areas for
Journal of Research in Science Teaching
572
CHANG ET AL.
developing structured educational interventions that increase the number of underrepresented
racial minority STEM degree recipients. In applying those interventions, institutions will need to
pursue them with a long-term commitment in ways that are mindful of the institution’s strengths
and weaknesses (Henderson, Beach, & Finkelstein, 2011). Fortunately, some of our recommended
interventions are already in place on some campuses, so a promising line of inquiry for future
research will be to more closely examine their educational efficacy, inclusiveness, and impact on a
broader range of students.
This research was supported in part by grants from the National Institute of General Medical
Sciences (R01 GMO71968-01 & R01 GMO71968-05), and the National Science Foundation (0757076).
Notes
1
Our response weights were calculated in two steps. In step 1, we used data from the National
Student Clearinghouse and institutional registrars to identify the students that did not complete at
least 4 years of higher education. We removed these students from the 2004 TFS data to make the
initial sample consist of only those students who persisted for at least 4 years. In step 2, we used the
persisting cohort of students and logistic regression to predict the probability of responding to the
CSS. Predictor variables came from the 2004 TFS, and included indicators of race, gender, high
school achievement, and reasons for attending college (a full list of variables in the model is
available upon request). We then used the coefficients from the significant predictors in the model
to calculate the probability that a student would respond to the CSS, and these response
probabilities were inverted to develop response weights. The general formula for developing a
non-response weight is: weight ¼ 1/(probability of response). After calculating response weights,
we compared the weighted and un-weighted samples from 2004 to determine whether our weights
inappropriately skewed any of the 2004 Freshman Survey variables. After confirming that the
weight had not adversely affected the distributions of variables from the 2004 Freshman Survey,
we created a final weight that was normalized to account for sample size. This was calculated by
dividing each student’s response weight by the average population response rate in order to avoid
inflating any statistics calculated using the weighted sample.
2
As described in Pampel (2000), the expected difference in predicted probabilities for
dichotomous predictors can be calculated as follows: (1) Compute the logged odds of the predicted
logit for the reference group, L0 ¼ ln(P0/(1 P0)); we used the mean of the outcome variable for
the whole sample as P0; (2) Compute the predicted logit for the keyed group, Lx, by adding L0 to
the parameter estimate for the predictor of interest, bx (Lx ¼ L0 þ bx); (3) Use the logit Lx to
compute the probability for the keyed group, Px ¼ 1=ð1 þ eLx Þ; (4) Subtract the difference in
probabilities, Px P0. For all non-dichotomous variables, we followed the same procedure but
substituted the mean of the variable ðxÞ for the reference group and x þ 1 for the keyed group.
3
We also found in our HGLM analyses (identical to the URM model, but not shown in this
paper) that the effect of selectivity was negative and significant on all students. For a 100-point
increase from the mean student body SAT score, the chances of persisting in a STEM field were
reduced by over 11 percentage points for all students, including White and Asian American
undergraduates. So, the negative effect of selectivity was not limited to just URM students but
tends to influence all students.
References
Adedokun, O. A., Bessenbacher, A. B., Parker, L. C., Kirkham, L. L., & Burgess, W. D. (2013). Research
skills and STEM undergraduate research students’ aspirations for research careers: Mediating effects of
research self-efficacy. Journal of Research in Science Teaching, 50, 940–951.
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
573
Allison, P. D. (2002). Missing data. Thousand Oaks, CA: Sage.
American Association for the Advancement of Science. (2001). In pursuit of a diverse science,
technology, engineering, and mathematics workforce: Recommended research priorities to enhance
participation by underrepresented minorities. Washington, DC: AAAS.
Anaya, G., & Cole, D. G. (2001). Latina/o student achievement: Exploring the influence of studentfaculty interactions on college grades. Journal of College Student Development, 42(1), 3–14.
Babbie, E. (2001). The Practice of Social Research. (9th ed.) Belmont, CA: Wadsworth/Thompson
Learning.
Bonous-Hammarth, M. (2000). Pathways to success: Affirming opportunities for science, mathematics,
and engineering majors. Journal of Negro Education, 69(1/2), 92–111.
Carlone, H. B., & Johnson, A. (2007). Understanding the science experiences of successful women of
color: Science identity as an analytic lens. Journal of Research in Science Teaching, 44(8), 1187–1218.
Chang, M. J., Cerna, O., Han, J., & Sáenz, V. (2008). The contradictory roles of institutional status in
retaining underrepresented minorities in biomedical and behavioral science majors. The Review of Higher
Education, 31(4), 433–464.
Chang, M. J., Eagan, M. K., Lin, M. H., & Hurtado, S. (2011). Considering the impact of racial stigma
and science identity: Persistence among biomedical and behavioral science students. Journal of Higher
Education, 82(5), 564–597.
Cole, D., & Espinoza, A. (2008). Examining the academic success of Latino students in science,
technology, engineering, and mathematics (STEM) majors. Journal of College Student Development, 49(4),
285–300.
Crisp, G., Nora, A., & Taggart, A. (2009). Student characteristics, pre-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.
Daempfle, P. A. (2003). An analysis of the high attrition rates among first year college science, math, and
engineering majors. Journal of College Student Retention, 5(1), 37–52.
DeAngelo, L., Franke, R., Hurtado, S., Pryor, J. H., & Tran, S. (2011). Completing college: Assessing
graduation rates at four-year institutions. Los Angeles: Higher Education Research Institute, UCLA.
de Leeuw, J., & Meijer, E. (2008). Introduction to multilevel analysis. In J. de Leeuw & E. Meijer (Eds.),
Handbook of multilevel analysis. (pp. 1–76). New York, NY: Springer.
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood from incomplete data via the
EM algorithm. Journal of the Royal Statistical Society, 39(1), 1–38.
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.
Espinosa, L. L. (2011). Pipelines and pathways: Women of color in undergraduate STEM majors and the
college experiences that contribute to persistence. Harvard Educational Review, 81(2), 209–240.
Gándara, P., & Maxwell-Jolly, J. (1999). Priming the pump: Strategies for increasing the achievement of
underrepresented minority undergraduates. New York, NY: The College Board.
Gasiewski, J. A., Eagan, M. K., Garcia, G. A., Hurtado, S., & Chang, M. J. (2012). From gatekeeping to
engagement: A multicontextual, mixed method study of student academic engagement in introductory
STEM courses. Research in Higher Education, 53(2), 229–261.
Grandy, J. (1998). Persistence in science of high-ability minority students. The Journal of Higher
Education, 69(6), 589–620.
Henderson, C., Beach, A., & Finkelstein, N. (2011). Facilitating change in undergraduate STEM
instructional practices: An analytic review of the literature. Journal of Research in Science Teaching, 48,
952–984.
Higher Education Research Institute. (2010). Degrees of success: Bachelor’s degree completion rates
among initial STEM majors. Los Angeles, CA: Higher Education Research Institute.
Huang, G., Taddese, N., & Walter, E. (2000). Entry and persistence of women and minorities in college
science and engineering education (No. NCES 2000601). Washington, DC: National Center for Education
Statistics.
Journal of Research in Science Teaching
574
CHANG ET AL.
Hurtado, S. (2007). The study of college impact. In P. Gumport (Ed.) The sociology of higher education:
Problems and prospects. Baltimore, MD: John Hopkins University Press.
Hurtado, S., Cabrera, N. L., Lin, M. H., Arellano, L., & Espinosa, L. (2009). Diversifying science:
Underrepresented student experiences in structured research programs. Research in Higher Education,
50(2), 189–214.
Hurtado, S., Eagan, M. K., Cabrera, N. L., Lin, M. H., Park, J., & Lopez, M. (2008). Training future
scientists: Predicting first-year minority student participation in health science research. Research in Higher
Education, 49(2), 126–152.
Hurtado, S., Han, J.C., Sáenz, V. B., Espinosa, L. L., Cabrera, N. L., & Cerna, O. S. (2007). Predicting
transition and adjustment to college: Biomedical and behavioral science aspirants’ and minority students’
first year of college. Research in Higher Education, 48(7), 841–887.
Kardash, C. M. (2000). Evaluation of an undergraduate research experience: Perceptions of
undergraduate interns and their faculty mentors. Journal of Educational Psychology, 92(1), 191–201.
Kinkead, J. (2003). Learning through inquiry: An overview of undergraduate research. In J. Kinkead
(Ed.). Valuing and supporting undergraduate research. New directions for teaching and learning, no. 93
(pp. 5–17). San Francisco: Jossey-Bass.
Liu, A., Ruiz, S., DeAngelo, L., & Pryor, J. (2009). Findings from the 2008 Administration of the
College Senior Survey (CSS): National aggregates. Los Angeles, CA: Higher Education Research Institute.
Accessed online on 10 April, 2010 at: http://www.heri.ucla.edu/PDFs/pubs/Reports/CSS2008_FinalReport.
pdf
Lopatto, D. (2003). The essential features of undergraduate research. Council on Undergraduate
Research Quarterly, 23, 139–142.
Lopez, E. J., Nandagopal, K., Shavelson, R. J., Szu, E., & Penn, J. (2013). Self-regulated learning study
strategies and academic performance in undergraduate organic chemistry: An investigation examining
ethnically diverse students. Journal of Research in Science Teaching, 50, 660–676.
Mabrouk, P. A., & Peters, K. (2000). Student perspectives on undergraduate research (UR) experiences
in chemistry and biology. Conferences on Chemistry. Retrieved October 25, 2006, from http://www.chem.vt.
edu/confchem/2000/a/mabrouk/mabrouk.htm
McLachlan, G. J., & Krishnan, T. (1997). The EM algorithm and extensions. New York, NY: Wiley.
Museus, S. D., Nichols, A. H., & Lambert, A. D. (2008). Racial differences in the effects of
campus racial climate on degree completion: A structural equation model. Review of Higher Education,
32(1), 107–134.
Mutegi, J. W. (2013). “Life’s first need is for us to be realistic” and other reasons for examining the
sociocultural construction of race in the science performance of African American students. Journal of
Research in Science Teaching, 50, 82–103.
National Academy of Sciences, National Academy of Engineering, & Institute of Medicine. (2007).
Rising above the gathering storm: Energizing and employing America for a brighter economic future.
Washington, DC: The National Academies Press.
National Science Foundation. (2009). Women, minorities, and persons with disabilities in science and
engineering. Arlington, VA: National Science Foundation.
National Research Council. (2011). Expanding Underrepresented Minority Participation: America’s
Science and Technology Talent at the Crossroads. Washington, DC: The National Academies Press.
Nora, A., Barlow, L., & Crisp, G. (2005). Student persistence and degree attainment beyond the first year
in college: The need for research. In A. Seidman (Ed.) College student retention: Formula for success.
Westport, CT: Praeger Publications.
Oseguera, L., Hurtado, S., Denson, N., Cerna, O., & Sáenz, V. (2006). The characteristics and
experiences of minority freshmen committed to biomedical and behavioral science research careers. Journal
of Women and Minorities in Science and Engineering, 12(2–3), 155–177.
Pampel, F. C. (2000). Logistic Regression: A primer. SAGE university paper series on quantitative
applications in the social sciences, 07-132. Thousand Oaks, CA: Sage.
Petersen, T. (1985). A comment on presenting results from logit and probit models. American
Sociological Review, 50(1), 130–131.
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
575
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis
methods. (2nd ed.). Thousand Oaks, CA: Sage.
Raudenbush, S. W., Bryk, A. S., Cheong, Y. F., & Congdon, R. (2004). HLM6: Hierarchical linear and
nonlinear modeling. Lincolnwood, IL: Scientific Software International, Inc.
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.
Sabatini, D. A. (1997). Teaching and research synergism: The undergraduate research experience.
Journal of Professional Issues in Engineering Education and Practice, 123(3), 98–102.
Sax, L. J., Hurtado, S., Lindholm, J. A., Astin, A. W., Korn, W. S., & Mahoney, K. M. (2004). The
American freshman: National norms for fall 2004. Los Angeles, CA: Higher Education Research Institute.
Seymour, E., & Hewitt, N. (1997). Talking about leaving: Why undergraduates leave the sciences.
Boulder, CO: Westview Press.
Sharkness, J., DeAngelo, L., & Pryor, J. (2010). CIRP construct technical report. Los Angeles, CA:
Higher Education Research Institute. Accessed online on 10 April 2010, at: http://www.heri.ucla.edu/PDFs/
technicalreport.pdf
Tinto, V. (1993). Leaving college: Rethinking the causes and cures of student attrition. (2nd ed.).
Chicago, IL: The University of Chicago Press.
Titus, M. A. (2004). An examination of the influence of institutional context on student persistence at
4-year colleges and universities: A multilevel approach. Research in Higher Education, 45(7), 673–699.
Titus, M. A. (2006). No college student left behind: The influence of financial aspects of a state’s higher
education policy on college completion. The Review of Higher Education, 29(3), 293–317.
Torres, J. B., & Solberg, V. S. (2001). Role of self-efficacy, stress, social integration, and family support
in Latino college student persistence and health. Journal of Vocational Behavior, 59(1), 53–63.
White, J. L., Altschuld, J. W., & Lee, Y. (2006). Persistence of interest in science, technology,
engineering, and mathematics: A minority retention study. Journal of Women and Minorities in Science and
Engineering, 12(1), 47–64.
Journal of Research in Science Teaching
576
CHANG ET AL.
Appendix A
Descriptive Statistics and Variable Coding (URM Students Only, Student N ¼ 1,634; Institution N ¼ 194)
Variables
Dependent variable
Persistence in a science, technology, engineering or math
major through fourth year of college (0 ¼ no, 1 ¼ yes)
Independent variables
Student background characteristics
Native American (0 ¼ no, 1 ¼ yes)
Latino (0 ¼ no, 1 ¼ yes)
Black/African American (0 ¼ no, 1 ¼ yes)
Student’s gender
35.4% Male (1)
64.6% Female (2)
Mother’s education
7.3% Grammar or less (1)
5.1% Some HS (2)
16.4% HS graduate (3)
4.1% Postsecondary (4)
18.0% Some college (5)
28.9% College grad (6)
3.2% Some grad school (7)
17.0% Grad degree (8)
Average high school grade
0.0% D (1)
0.6% C (2)
1.0% Cþ (3)
3.1% B (4)
11.6% B (5)
17.1% Bþ (6)
29.0% A (7)
37.6% A OR Aþ (8)
Math þ Verbal SAT Score (in 100-point increments)
Years of mathematics in high school
0.0% None (1)
0.1% 1/2 year (2)
0.5% 1 year (3)
1.1% 2 years (4)
8.5% 3 years (5)
77.4% 4 years (6)
12.3% 5þ years (7)
Years of physical science in high school
6.1% None (1)
2.7% 1/2 year (2)
28.8% 1 year (3)
34.4% 2 years (4)
16.4% 3 years (5)
10.5% 4 years (6)
1.0% 5þ years (7)
Years of biological science in high school
1.9% None (1)
2.0% 1/2 year (2)
43.5% 1 year (3)
34.6% 2 years (4)
10.0% 3 years (5)
6.7% 4 years (6)
1.3% 5þ years (7)
Participated in Summer Research Program
85.7% No (1)
14.3% Yes (2)
Mean
S.D.
Min.
Max.
0.58
0.49
0
1
0.12
0.50
0.38
1.65
0.33
0.50
0.49
0.48
0
0
0
1
1
1
1
2
5.05
2.07
1
8
6.81
1.25
2
8
11.44
5.99
1.75
0.57
6.1
2
3.88
1.27
1
7
3.74
1.05
1
7
1.14
0.35
1
2
16
7
(Continued)
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
Variables
Intend to obtain Master’s degree (M.A., M.S., etc.)
(0 ¼ no, 1 ¼ yes)
Intend to obtain Ph.D. or Ed.D. (0 ¼ no, 1 ¼ yes)
Intend to obtain M.D., D.O., D.D.S., D.V.M. (0 ¼ no,
1 ¼ yes)
Entering science identity
TFS academic self-concept
TFS social self-concept
Concern about financing college education
21.9% None (1)
59.7% Some (2)
18.4% Major (3)
College experiences
Worked full-time while attending school
78.3% No (1)
21.7% Yes (2)
Felt family support to succeed
7.6% Not at all (1)
30.9% Occasionally (2)
61.6% Frequently (3)
Faculty mentoring factor
Asked a professor for advice outside of class
19.4% Not at all (1)
59.3% Occasionally (2)
21.2% Frequently (3)
Felt intimidated by your professors
42.4% Not at all (1)
49.3% Occasionally (2)
8.3% Frequently (3)
Faculty feel that most students here are well-prepared
academically
1.9% Strongly disagree (1)
17.5% Disagree (2)
65.0% Agree (3)
15.7% Strongly agree (4)
Studied with other students
4.7% Not at all (1)
47.0% Occasionally (2)
48.2% Frequently (3)
Sense of belonging
Positive cross racial interaction
Negative cross racial interaction
Hostile racial climate
There is strong competition among most of the students
for high grades
3.3% Strongly disagree (1)
27.8% Disagree (2)
44.4% Agree (3)
24.5% Strongly agree (4)
Felt overwhelmed by all I had to do
6.8% Not at all (1)
59.3% Occasionally (2)
33.8% Frequently (3)
Participated in an undergraduate research program (e.g.
MARC, etc.)
79.3% No (1)
20.7% Yes (2)
Joined a club or organization related to your major
39.9% No (1)
60.1% Yes (2)
577
Mean
S.D.
Min.
0.22
0.41
0
Max.
1
0.29
0.29
0.45
0.45
0
0
1
1
0.03
51.86
48.35
1.97
0.85
7.8
9.35
0.63
1.94
23.86
18.06
1
1.22
0.41
1
2
2.54
0.63
1
3
0.03
2.01
0.95
0.64
2.01
1
1.66
0.63
1
3
2.94
0.63
1
4
2.43
0.58
1
3
0.02
0.13
0.07
0.16
2.90
0.92
0.91
0.88
0.85
0.80
3.17
2.6
1.01
1.31
1
2.27
0.58
1
3
1.21
0.41
0.96
2
1.60
0.49
1
2
1.86
66.92
68.14
3
1.6
3
1.35
1.4
2.97
2.59
4
(Continued)
Journal of Research in Science Teaching
578
CHANG ET AL.
Variables
Worked on a professor’s research project
60.4% Not at all (1)
27.5% Occasionally (2)
12.1% Frequently (3)
Institution-level variables
Institutional control
43.8% Public (1)
56.2% Private (2)
Institutional Type
39.2% University (1)
60.8% Four-year (2)
HBCU
90.7% No (1)
9.3% Yes (2)
Selectivity (in 100-point increments)
Percent of students on financial aid in 2006 (in 10-point
increments)
Percent of students receiving federal aid in 2006
(in 10-point increments)
Any research expenditures in 2006
Percent of students majoring in STEM in 2006
(in 10-point increments)
Percent of student body that is American Indian, Black
or Latino in 2006
Log (Undergraduate FTE in 2006)
Mean
S.D.
Min.
1.51
0.70
1
Max.
3
1.56
0.50
1
2
1.61
0.49
1
2
1.09
0.29
1
2
11.15
8.01
1.48
1.71
7.8
0
2.78
1.81
0
9.5
0.81
1.66
0.39
1.50
0
0
1
8.9
2.57
2.50
0.28
9.94
8.62
0.94
6
15.1
10
10.51
Source: Cooperative Institutional Research Program 2004 Freshman Survey, 2008 College Senior Survey, and 2006
Integrated Postsecondary Data System.
a
Indicates multi-item factor, see Appendix B for more details.
Journal of Research in Science Teaching
RETAINING ASPIRING SCIENTISTS
579
Appendix B
Multi-Item Factors
Scale & Items
Entering Science Identity (Freshman Year)a
Become authority in my own field
Obtain recognition from colleagues
Make theoretical contribution to science
Work to find cure for health problem
a
All items on a 4-point scale, 1 ¼ not important,
4 ¼ essential
Faculty mentoringb
Encouragement to pursue graduate/professional study
An opportunity to work on a research project
Advice and guidance about your educational program
Emotional support and encouragement
A letter of recommendation
Help to improve your study skills
Feedback about your academic work (outside of grades)
An opportunity to discuss coursework outside of class
Help in achieving your professional goals
b
All items on a 3-point scale, 1 ¼ not at all, 3 ¼ frequently
Sense of belongingc
I see myself as part of the campus community
I feel I am a member of this college
I feel I have a sense of belonging to this campus
c
All items on a 4-point scale, 1 ¼ strongly disagree,
4 ¼ strongly agree
Positive cross-racial interactiond
Dined or shared a meal
Had meaningful and honest discussions about racial/ethnic
relations
Shared personal feelings and problems
Had intellectual discussions outside of class
Studied or prepared for class
Socialized or partied
Attended events sponsored by other racial/ethnic groups
d
All items on a 5-point scale, 1 ¼ never, 5 ¼ very often
Negative cross-racial interactione
Had guarded interactions
Had tense, somewhat hostile interactions
Felt insulted or threatened because of your race/ethnicity
e
All items on a 5-point scale, 1 ¼ never, 5 ¼ very often
Hostile Racial Climatef
I have been singled out because of my race/ethnicity,
gender, or sexual
I have heard faculty express stereotypes about racial/ethnic
groups in class
There is a lot of racial tension on this campus
f
All items on a 4-point scale, 1 ¼ strongly disagree,
4 ¼ strongly agree
All Students
Factor Loadings
URM Students
Factor Loadings
Alpha 0.673
0.590
0.694
0.579
0.488
Alpha 0.653
0.546
0.632
0.586
0.510
Alpha 0.896
0.692
0.590
0.787
0.747
0.644
0.651
0.724
0.656
0.825
Alpha 0.894
0.684
0.567
0.776
0.753
0.628
0.665
0.722
0.651
0.828
Alpha 0.879
0.790
0.855
0.884
Alpha .882
.786
.872
0.881
Alpha 0.897
0.779
0.762
Alpha 0.891
0.746
0.758
0.825
0.814
0.710
0.730
0.618
0.794
0.802
0.685
0.725
0.646
Alpha 0.771
0.661
0.810
0.721
Alpha 0.762
0.670
0.795
0.701
Alpha 0.682
0.698
Alpha 0.690
0.728
0.657
0.623
0.586
0.610
Source: Cooperative Institutional Research Program 2004 Freshman Survey and 2008 College Senior Surveys
Note: The multi-item factors TFS Academic Self-Concept and Social Self-Concept were created by researchers at CIRP
using Item Response Theory (IRT). Academic Self-Concept was constructed using items representing students’ selfratings of their Academic ability, Drive to achieve, Mathematical ability, and Self-confidence (intellectual). Social SelfConcept was constructed using items representing students’ self-ratings of their Leadership; Public speaking ability,
Self-confidence (social), and Popularity. See Sharkness, DeAngelo & Pryor (2010) and http://www.heri.ucla.edu/PDFs/
constructs/Appendix2009.pdf for more details.
Journal of Research in Science Teaching
Appendix C
Journal of Research in Science Teaching
group for race variables is White/Asian.
Reference group for degree aspiration variables is Bachelor’s degree only.
a
Reference
b
Demographic characteristics
Native Americana
Latinoa
Blacka
Student’s gender
Mother’s education
High school academic preparation
Average high school grade
Math þ Verbal SAT score (in 100-point increments)
Mathematics
Physical science
Biological science
Participated in Summer Research Program
Other pre-college characteristics
Intend to obtain Master’s Degreeb
Intend to obtain Ph.D. or Ed.D.b
Intend to obtain M.D., D.O., D.D.S., D.V.M.b
Entering science identity
TFS academic self-concept
TFS social self-concept
Concern about financing college education
College experiences
Worked full-time while attending school
Felt family support to succeed
Faculty mentoring factor
Asked a professor for advice outside of class
Felt intimidated by your professors
Faculty feel that most students here are well-prepared academically
Studied with other students
Sense of belonging
Positive cross racial interaction
Negative cross racial interaction
Hostile racial climate
There is strong competition among most of the students for high grades
Felt overwhelmed by all I had to do
Participated in an undergraduate research program (e.g., MARC, etc.)
Joined a club or organization related to your major
Worked on a professor’s research project
Intercept
0.17
0.08
0.10
0.07
0.02
0.13
1.25
SE
0.14
0.25
0.39
0.35
0.07
B
9.75
0.83
3.03
3.80
4.73
3.35
T
Model 1
0.00
0.41
0.00
0.00
0.00
0.00
p
0.11
0.13
0.12
0.05
0.01
0.00
0.07
0.30
0.15
0.13
0.13
0.05
0.03
0.00
0.15
0.04
0.03
0.08
0.04
0.04
0.12
0.12
0.18
0.13
0.05
0.00
0.11
0.99
0.19
0.09
0.12
0.09
0.02
SE
0.04
0.02
0.04
0.20
0.02
B
6.46
2.63
1.21
1.08
2.51
6.25
7.51
0.04
3.21
5.57
1.58
1.51
0.12
0.86
0.21
0.17
0.35
2.26
0.85
T
Model 2
0.00
0.01
0.23
0.28
0.01
0.00
0.00
0.97
0.00
0.00
0.11
0.13
0.90
0.39
0.84
0.87
0.73
0.02
0.39
p
0.51
0.07
0.12
0.20
0.07
0.04
0.49
0.02
0.03
0.14
0.15
0.11
0.12
0.72
0.39
0.36
1.27
0.03
0.18
0.25
0.34
0.05
B
0.10
0.07
0.06
0.07
0.07
0.08
0.07
0.06
0.05
0.07
0.06
0.06
0.08
0.12
0.09
0.07
0.15
0.19
0.09
0.13
0.09
0.02
SE
4.92
1.10
2.12
2.89
1.00
0.54
7.22
0.27
0.59
2.22
2.59
1.92
1.56
6.09
4.26
5.49
8.40
0.17
1.93
1.94
3.97
2.06
T
Model 3
0.00
0.27
0.03
0.00
0.32
0.59
0.00
0.79
0.56
0.03
0.01
0.06
0.12
0.00
0.00
0.00
0.00
0.86
0.05
0.05
0.00
0.04
p
Hierarchical Generalized Linear Modeling (HGLM) Results for all students (N ¼ 3,670) persistence in a Science, Technology, Engineering or Math (STEM) major
580
CHANG ET AL.