Académique Documents
Professionnel Documents
Culture Documents
Spring 2017
Recommended Citation
Rickels, Heather Anne. "Predicting college readiness in STEM: a longitudinal study of Iowa students." PhD
(Doctor of Philosophy) thesis, University of Iowa, 2017.
https://doi.org/10.17077/etd.dcit9d2l
by
May 2017
2017
CERTIFICATE OF APPROVAL
____________________________
PH.D. THESIS
_________________
____________________________________________
Stephen Dunbar
____________________________________________
Brandon LeBeau
____________________________________________
Megan Foley Nicpon
____________________________________________
Kyong Mi Choi
To my Anthonys
ii
ACKNOWLEDGEMENTS
This dissertation would not have been possible without support from many
people. Foremost, I would like to thank my advisor and committee chair, Dr. Catherine
Welch for invaluable guidance. It was your encouragement that first led me to apply to
the PhD program. Your insightful feedback throughout my time here has always been
provided with enthusiasm and kindness. I am also grateful for the guidance and support
of Dr. Stephen Dunbar. Your test construction class sparked my interest in our field and
you continued to foster this interest through several statistics courses. Thank you for your
statistical methods and reporting results. I would also like to express sincere appreciation
for the other members of my committee. Dr. Brandon LeBeau, thank you for your
attention to technical details and for your suggestions related to modifying the methods.
Dr. Kyong Mi Choi, thank you for your fresh perspective on STEM and your thought
provoking feedback. And Dr. Megan Foley Nicpon, I am grateful for your support
Several others also aided in making this dissertation possible. Dr. Matthew
Whittaker provided unwavering support during my time at Iowa Testing Programs and
assembled the datasets for this dissertation. Conversations with Dr. Paul Westrick incited
enthusiasm for this topic. In addition, he sent numerous key articles and resources my
way during the planning stages of this dissertation. In addition, I greatly appreciate Dr.
iii
Anthony Fina, Dr. Angelica Rankin, Dr. Rebecca Riley, and Jennie Rickels for
Finally, thank you to my family and friends for their love and support throughout
this journey. A very special thanks to Anthony for your support, beginning with
providing feedback on my first ideas for the dissertation to editing the final drafts. Thank
you for debating details that no one else would have understood or cared about. I am
forever grateful for your never-ending support and encouragement. I truly could not have
done this without you. In addition, thank you to the many friends I have made during my
time in graduate school, especially Rebecca, Lauren, Angelica, Charles, Aaron, Wei
Cheng, Joleen, and Barbara. Last, but certainly not least, thank you to my mom for
instilling a love of learning, for always believing in me, and for making me believe that I
iv
ABSTRACT
The demand for STEM college graduates is increasing. However, recent studies
show there are not enough STEM majors to fulfill this need. This deficiency can be
partially attributed to a gender discrepancy in the number of female STEM graduates and
to the high rate of attrition of STEM majors. As STEM attrition has been associated with
STEM graduates change in achievement levels from middle school through high school
and to have accurate readiness indicators for first-year STEM coursework. This study
aimed to address these issues by comparing the achievement growth of STEM majors to
non-STEM majors by gender in Science, Math, and Reading from Grade 6 to Grade 11
through latent growth models (LGMs). Then STEM Readiness Benchmarks were
established in Science and Math on the Iowas (IAs) for typical first-year STEM courses
Results from the LGM analyses demonstrated that STEM graduates start at higher
achievement levels in Grade 6 and maintain higher achievement levels through Grade 11 in
all three subjects studied. In addition, gender differences were examined. The findings
indicate that students with high achievement levels self-select as STEM majors, regardless of
gender. They also suggest that students who are not on-track for a STEM degree may need to
begin remediation prior to high school. Results from the benchmark analyses indicate that
STEM coursework is more demanding and that students need to be better prepared
academically in science and math if planning to pursue a STEM degree. The STEM
v
courses than if general college readiness benchmarks were utilized. Also, students who
met the STEM Readiness Benchmarks were more likely to graduate with a STEM degree.
educators, and college admissions officers. Findings from this study can be used to better
major and to provide guidance for students considering STEM majors in college. If
students are being encouraged to purse STEM majors, it is important they have accurate
vi
PUBLIC ABSTRACT
The demand for STEM college graduates is increasing. However, recent studies
suggest there are not enough STEM majors to fulfill this need. This deficiency can be
partially attributed to a lack of female STEM graduates and to the high rate of attrition of
STEM majors. As STEM attrition has been associated with students being unprepared for
successful as a STEM graduate. This study addressed these issues through comparing the
achievement growth of STEM majors to non-STEM majors from middle school to high
school. Then cut scores were established on a state achievement test that indicate whether
students are prepared for typical first-year STEM courses. These cut scores are called
Results from the analyses demonstrated that STEM graduates start at higher
achievement levels in Grade 6 and maintain higher achievement levels through Grade
11. In addition, gender differences were considered. The findings indicate that students
with high achievement levels self-select as STEM majors, regardless of gender. They
also suggest that students who are not on-track for a STEM degree may need to begin
remediation prior to high school. Finally, results from the STEM Readiness Benchmark
analyses indicate that STEM coursework is more demanding and that students need to be
better prepared academically in science and math if planning to pursue a STEM degree.
Overall, this study provides valuable information on STEM readiness to students and
educators. If students are being encouraged to purse STEM majors, it is important they
vii
TABLE OF CONTENTS
STEM ...................................................................................................................................................... 8
Majoring in STEM .......................................................................................................................... 9
Leaky STEM Pipeline.................................................................................................................. 10
Women in STEM .......................................................................................................................... 13
Contemporary STEM Initiatives ................................................................................................... 17
College Readiness.............................................................................................................................. 20
Predicting College Readiness .................................................................................................... 21
College Readiness Benchmarks ................................................................................................ 21
STEM Readiness Benchmarks .................................................................................................. 22
Measuring Achievement Growth ............................................................................................. 24
Latent Growth Models...................................................................................................................... 28
Summary .............................................................................................................................................. 30
Background .......................................................................................................................................... 33
Data Sources ........................................................................................................................................ 34
Iowa Testing Programs Archival Data .................................................................................... 34
University of Iowa Registrar Records ..................................................................................... 35
Matching Student Records ......................................................................................................... 36
Study Procedures ............................................................................................................................... 37
Objective 1: Comparing Achievement Growth of STEM and Non-STEM Majors 38
Objective 2: Establishing STEM Readiness Benchmarks on the IAs............................ 43
Objective 3: Provide Validity Evidence for the STEM Readiness Benchmarks ...... 45
Summary .............................................................................................................................................. 46
viii
Rationale for Methods ................................................................................................................. 54
Science ............................................................................................................................................. 55
Math .................................................................................................................................................. 58
Reading ............................................................................................................................................ 60
Covariates: Gender and STEM ................................................................................................. 62
Summary.......................................................................................................................................... 64
Objective 2: Establishing STEM Readiness Benchmarks on the IAs ................................ 65
Course Selection ............................................................................................................................ 65
Descriptive Statistics .................................................................................................................... 66
Science ............................................................................................................................................. 67
Math .................................................................................................................................................. 70
Summary.......................................................................................................................................... 71
Objective 3: Provide Validity Evidence for the STEM Readiness Benchmarks ............. 72
Error Rates ...................................................................................................................................... 72
Chi-Square Goodness of Fit Tests............................................................................................ 75
Summary.......................................................................................................................................... 81
Summary .............................................................................................................................................. 81
ix
LIST OF TABLES
Table 2.1 Summary of the Distribution of Students’ First College Math and
Science Courses by STEM Category……………..……………...….........32
Table 4.2 Means (SDs) of Scores on the Iowas for LGM Datasets….....…..….…....84
Table 4.3 Model Estimates for the Latent Growth Models in Science, Math,
and Reading….....………….…………………………………………......85
Table 4.4 Fit Statistics for the Latent Growth Models in Science, Math, and
Reading…..........................................................................................…....85
Table 4.5 Observed and Estimated Means for the Latent Growth Models in
Science, Math, and Reading….....………….………………………….....86
Table 4.6 Covariate Loadings for the Conditional Latent Growth Models in
Science, Math, and Reading….....………….………………………….....86
Table 4.8 Means (SDs) of Scores on the Iowas for STEM Benchmark Datasets…..87
Table 4.12 Values for the STEM Readiness Benchmarks versus General CRBs……89
x
Table 4.14 Cumulative Percent Frequency Distribution of Grades in Selected Math
Courses ….....……………………………………………………….…....90
Table 4.15 Error Rates for the STEM Readiness Benchmarks versus General
CRBs……………………………………………….........………….…....91
Table 4.16 Error Rates for the Science STEM Readiness Benchmark by Science
Course..…………………………………………….........………….…....91
Table 4.17 Error Rates for the Math STEM Readiness Benchmark by Selected
Math Course..……...……………………………….........………….…....91
Table 4.19 Means (SDs) of Scores on the Iowas for the Science STEM Readiness
Benchmark Dataset………..……………………….........………….…....93
Table 4.20 Means (SDs) of Scores on the Iowas for the Math STEM Readiness
Benchmark Dataset..……...….…………………….........………….…....94
Table 4.21 Means (SDs) of ACT Scores for STEM Readiness Benchmark
Datasets..……...…………………………...……….........………….…....95
xi
LIST OF FIGURES
xii
CHAPTER I: INTRODUCTION
There has been an increased focus on science and math achievement in the
United States due to a concern that students lack the skills needed for careers in science,
technology, engineering, and mathematics (STEM) fields (National Center for Education
Statistics, 2011). While the demand for STEM college graduates is increasing, there are
not enough graduates to fulfill this need (e.g., U.S. Department of Commerce, 2011a;
U.S. Department of Labor, 2007). Given this fact, many find it alarming that of the 2015
high school graduates who took the ACT, only 38% were ready for college level
coursework in science and only 42% were ready for college level coursework in
[CoSTEM], 2013). The STEM Strategic Plan outlined five goals, including 1) improving
traineeships. As a response, there has been a surge of research in areas such as gender
differences in STEM, selecting a STEM major, and persisting in a STEM major (e.g.,
1
As women are proportionally underrepresented in STEM (U.S. Department of
appears in both the fact that women are less likely to major in STEM fields (Heilbronner,
2013), and also that women are less likely to work in STEM jobs after obtaining a STEM
degree (U.S. Department of Commerce, 2011a). For example, in 2013, women earned
only 36% of all STEM undergraduate degrees (American Physical Society, 2015).
Furthermore, while 40% of men with STEM degrees work in STEM fields, only 26% of
women with STEM degrees work in STEM fields (U.S. Department of Commerce,
2011a). It is important to note that these gender differences cannot be accounted for by
prior achievement in science and mathematics (Riegle-Crumb, Kind, Grodsky, & Muller,
2012) or by achievement in college as a STEM major (Huang, Taddese, & Walter, 2000).
In general, researchers have found that both males and females are more likely to
major in STEM if they have higher levels of interest (ACT, 2014b; Tai, Liu, Maltese, &
Fan, 2006) and higher levels of high school achievement as measured by courses taken,
grade point average, and standardized test scores (e.g., LeBeau et al., 2012; Wang, 2013).
However, while 49% of high school graduates who take the ACT express interest in
majoring in STEM (ACT, 2014b), a much smaller percentage actually declare a STEM
[PCAST], 2012). In addition, approximately 50% of students who declare a STEM major
do not complete the degree (Chen, 2013; PCAST, 2012). STEM attrition is a term used to
refer to this situation in which college students declare a STEM major, but then do not
2
graduate with a STEM degree due to either switching to a non-STEM major or leaving
STEM attrition is a major concern in the field of STEM because of the need for
more STEM majors, as well as the consequences for some students, such as dropping out
of college (Chen, 2013; Stinebrickner & Stinebrickner, 2014; Westrick, 2015). STEM
attrition has been attributed to performing poorly in STEM classes (Stinebrickner &
Stinebrickner, 2014; Wang, 2013). In addition, leaving college without a degree has been
associated with poor performance in college and higher levels of withdrawn or failed
STEM classes (Chen, 2013). Alternately, successful STEM majors tend to demonstrate
high school grade point average (e.g., Chen, 2013; LeBeau et al., 2012; Westrick, 2015).
As high school students are being encouraged to consider STEM majors, it is important
to ensure that these students are academically prepared for the rigors of a STEM major.
STEM Readiness
Surprisingly, very little research has specifically examined the college readiness
of STEM majors. Given the fact that STEM courses tend to be more academically
challenging than non-STEM courses (Ost, 2010; Phillip, Brennan, & Meleties, 2005;
Rask, 2010; Stinebrickner & Stinebrickner, 2014), it follows that general college
readiness indicators may not be suitable for indicating a student’s readiness for STEM
validating STEM Readiness Benchmarks (Mattern, Radunzel, & Westrick, 2015). These
college readiness benchmarks are higher than ACT’s general College Readiness
3
important first step, there is still a lack of research linking grade school achievement to
college outcomes of STEM majors. This void is especially troubling as research suggests
that college and career readiness planning should begin before high school (ACT, 2008).
The goal of this dissertation is to address this void by comparing the growth of
STEM majors and non-STEM majors on the ITBS (Hoover, Dunbar, & Frisbee, 2001)
and the ITED (Forsyth, Ansley, Feldt, & Alnot, 2001), which will collectively be referred
to as the Iowas (IAs), starting in Grade 6. Growth will be modeled through latent growth
models (LGMs), which offer a flexible way to investigate the nature of growth over time.
This dissertation will also establish STEM Readiness Benchmarks on the IAs in order to
provide students with a more accurate picture of their current preparedness for college
coursework required for STEM majors. Research findings suggest that being more
prepared may help decrease STEM attrition (e.g., Chen, 2013; LeBeau et al., 2012;
Westrick, 2015). Therefore, similar to the IAs’ CRBs established by Fina (2014), these
STEM benchmarks will allow for remediation if students are interested in STEM careers,
but do not currently meet the new benchmarks. In addition, the STEM benchmarks may
Statement of Purpose
There is a call for increasing the number of STEM graduates in the United States
(CoSTEM, 2013); however, there is not a clear consensus on how to achieve this goal. As
mentioned above, possible solutions include increasing STEM interest, increasing STEM
education, and finding ways to retain current STEM majors. However, researchers have
found that students who are less academically prepared are not only
4
more likely to switch majors, but also more likely to quit college entirely (Chen, 2013;
PCAST, 2012; Stinebrickner & Stinebrickner, 2014; Westrick, 2015). This suggests that
prior to college. As STEM courses have been shown to be more demanding than general
education courses, it follows that they may require a higher level of achievement needed
to be considered college ready in these fields. This dissertation strives to provide more
Iowa. In order to accomplish this, the three research objectives outlined below will be
examined:
5
3. Examine and evaluate the validity evidence for the STEM Readiness
b. Calculate the error rates and PCC for selected individual Science and
Math courses. Compare these rates with those found using all courses
differently by course.
c. Calculate the error rates and PCC for the STEM Readiness
major), and growth from Grades 6 to 11. Compare these rates through
Summary
This study will compare the achievement growth and college readiness of STEM
established on the IAs using grades from science and math courses typically taken by
STEM majors during the first year of college. The validity evidence for these
graduating with a STEM degree will then be examined. Together, these three objectives
6
successfully graduate from a STEM field of study in college. Chapter II will discuss the
STEM, and measurement of student achievement growth. Chapter III will describe the
research methodology used to address the specific objectives. Chapter IV will report the
results of the study. Chapter V will summarize the main findings, and then discuss the
7
CHAPTER II: LITERATURE REVIEW
Chapter I outlined the foundational concepts for this dissertation. Chapter II provides
a review of the current literature on STEM, including themes related to selecting a STEM
major, STEM attrition, and women in STEM. Contemporary educational initiatives as they
relate to STEM education and college readiness are also discussed and the need for STEM
Readiness Benchmarks is highlighted. In addition, the history of the Iowas (IAs) as indicators
of growth towards college readiness are summarized. Lastly, the latent growth models
STEM
There has been an increased focus on science and math achievement in the
United States. This is, in part, because the demand for STEM graduates is increasing,
while research shows there is a lack of STEM graduates to meet this need (e.g., U.S.
Department of Commerce, 2011a; U.S. Department of Labor, 2007). STEM jobs are
growing faster than those in other fields (U.S. Department of Labor, 2007). By 2018, it is
estimated that 9 out of the 10 fastest growing jobs requiring at least a college degree will
be in a STEM field (Lacey & Wright, 2009). Given these statistics, many find it alarming
that of the 2015 graduates who took the ACT, only 38% were ready for college level
coursework in science and 42% were ready for college level coursework in mathematics
(ACT, 2015). The picture is only slightly better for Iowa students; of the 67% of 2015
graduates who took the ACT, 48% were ready for college level coursework in science
and 48% were ready for college level coursework in mathematics (ACT, 2015). These
college-ready statistics were very similar to those from 2014, both at the
8
state and national level (ACT, 2014a). Further complicating this scenario is the fact that
many students who are talented in STEM do not choose to pursue a STEM major
(Heilbronner, 2009), even though 49% of all 2014 graduates who took the ACT
There are many reasons choosing a STEM major would be advantageous. For
example, STEM fields are growing and jobs are more readily available than in other
average of 26% more than their counterparts in non-STEM jobs; this is true across all
levels of degree attainment, from high school diploma to doctorate degree (U.S.
Department of Commerce, 2011a). Furthermore, STEM majors tend to earn more even
when they are not working in STEM occupations (U.S. Department of Commerce,
2011a). Given these economic benefits, it is especially surprising that more college
Majoring in STEM
Many researchers have examined factors influencing the selection of a STEM major
and attaining a STEM degree in college. It has been found that exposure to math and science
courses before college, a high degree of self-efficacy in STEM, level of high school math
directly influence this decision (Wang, 2013). Heilbronner (2009) also found that a student’s
self-efficacy in STEM and the quality of a student’s academic experiences in math and
science courses influenced talented individuals to select a STEM major. LeBeau et al. (2012)
examined the relationship between the high school math curricula students are exposed to
9
found that high school math curriculum was not a significant predictor, but ACT math
score, gender, and high school math GPA were significant (LeBeau et al., 2012).
Interestingly, when eighth-grade students expressed intent to major in STEM, they were
3.4 times more likely to earn a degree in physical science or engineering than students
who did not express an interest (Tai et al., 2006). This suggests that early exposure to
STEM, precollege STEM achievement, and student interest in STEM can all affect future
STEM involvement.
Strategies to increase the number of STEM majors have also been researched.
For example, Heidel et al. (2011) found that support for walk-in tutoring, early
undergraduate research experiences, and adult learner scholarships were all successful in
motivation and performance in STEM courses (Chang, Sharkness, Hurtado, & Newman,
2014; Lloyd & Eckhardt, 2010). Watkins and Mazur (2013) found that utilizing an
who declared STEM majors was not found to be effective in increasing the number of
STEM graduates (Heidel et al., 2011). These students did not tend to persist as STEM
concerns are often discussed: the leaky STEM pipeline and STEM attrition. The leaky
STEM pipeline refers to the fact that potential STEM majors are lost at many stages
10
along the pathway to becoming employed in STEM occupations. STEM attrition is one
such leak in the pipeline. STEM attrition specifically refers to students who initially
declare a STEM major, but do not obtain a STEM degree because they either switched to
(PCAST, 2012).
It has been found that students leave the STEM pipeline at various points.
Students enter college equally open to majoring in STEM as compared to other fields
(Stinebrickner & Stinebrickner, 2014). However, while 49% of students express interest
in majoring in STEM (ACT, 2014b), a much smaller percentage actually declare a STEM
major and an even smaller percentage earn a STEM degree (Chen, 2013; PCAST, 2012).
This is troubling as most STEM jobs require a bachelor’s degree (Carnevale, Smith, &
Melton, 2011). Analyses on a national sample of college students found that only 28%
declared a STEM major at some point, and 48% of those students did not graduate with a
STEM major (Chen, 2013). Of the students who declared but did not earn a STEM
degree, approximately half switched to a non-STEM major, and the other half left college
before obtaining any degree (Chen, 2013). Students are more likely to leave STEM
majors than other majors, and are less likely to switch to STEM majors from other
majors (Stinebrickner & Stinebrickner, 2014). Furthermore, only about 56% of students
with STEM majors end up working in STEM-related fields (Carnevale et al., 2011).
call for increasing the number of STEM graduates in the United States by 1 million in the
next ten years (CoSTEM, 2013). Given this priority to increase the number of STEM
graduates, it is highly disconcerting to lose such large portions of STEM majors along the
11
pipeline. STEM researchers universally acknowledge that we need to stop the leak (e.g.,
Chen, 2013; Mattern et al., 2015; PCAST, 2012). The President’s Council of Advisors on
Science and Technology (PCAST) argues that retaining STEM majors is the fastest way
to attain the additional 1 million STEM majors our workforce will require in the next
decade. Specifically, PCAST cites the fact that only 40% of students intending to major
in STEM do so, but if 50% of them did instead, this would result in three-quarters of the
STEM attrition is a concern, not only because of the need for more STEM majors,
but also because of the consequences for students, such as dropping out of college (Chen,
2013; Stinebrickner & Stinebrickner, 2014; Westrick, 2015). STEM attrition has been
associated with performing poorly in STEM classes (Chen, 2013; Wang, 2013).
Stinebrickner and Stinebrickner (2014) found that the first STEM classes have a large
impact on whether students persist in a STEM major. In particular, students often leave
science majors because they are performing worse in STEM classes than expected. They
also found that students attribute this poor performance to ability rather than willingness
to study. Chen (2013) specifically compared factors associated with the two types of
STEM attrition: switching to non-STEM major and leaving college. Switching to non-
STEM majors was associated with taking fewer STEM courses during the first year and
performing poorly in STEM classes relative to non-STEM classes (Chen, 2013). Leaving
college without a degree was associated with poor overall performance in college and
higher numbers of withdrawn or failed STEM classes (Chen, 2013). Additionally, low
performing students in STEM majors (i.e., overall GPA less than 2.5) were more likely to
drop out of college while high performing STEM majors (i.e., overall GPA greater than
12
3.5) were more likely to switch to other majors. Westrick (2015) completed a meta-
STEM majors, switched to STEM majors, and left STEM majors. He found that those
who persisted in a STEM major or switched to a STEM major had higher ACT scores
and high school grade point averages (HSGPAs) than those who left, but similar levels of
STEM interest. This suggests that while STEM interest is associated with obtaining a
STEM degree, interest alone is not enough for students to be successful in STEM.
ensure that they are academically prepared for the rigors of a STEM major. This
sentiment cannot be understated and has been echoed by other researchers. For example,
Stinebrickner and Stinebrickner (2014) concluded that students are choosing science
majors based on their abilities, so in order to increase science majors, we need to better
prepare students for the demands of college level science. Also, Mattern et al. (2015)
concluded that we “must seek ways to maintain the STEM pipeline with students who
are likely to succeed in a STEM major and persist in a STEM field” (p. 2). Students are
more likely to succeed when they possess high levels of academic achievement before
Women in STEM
U.S. Department of Commerce (2011a), “Half as many women are working in STEM
jobs as one might expect if gender representation in STEM professions mirrored the
13
overall workforce” (p. 2). This is especially poignant given the large need for more
One issue is that women are less likely to major in STEM fields (Heilbronner,
2013). The percentage of women earning STEM degrees has increased in the past fifty
years, however, in 2013, women still earned only 36% of all STEM undergraduate
degrees (American Physical Society, 2015). Not only are women less likely to major in
STEM fields, they are less likely to work in STEM jobs even if they do major in STEM
fields (Heilbronner, 2013). While 40% of men with STEM degrees work in STEM
fields, only 26% of women with STEM degrees work in STEM fields (U.S. Department
have historically been less likely to earn a STEM Ph.D. (Alper, 1993; Zwick, 1991).
However, new research suggests that the gender gap in the persistence rates for STEM
in math and science (Riegle-Crumb et al., 2012). In high school, males and females earn
equal number of credits in math and science, although females earn slightly higher grades
(U.S. Department of Education, National Center for Educational Statistics, 2007). While
small gender gaps do remain on some national standardized tests (Quinn & Cooc, 2015),
they explain very little of the gender gap in STEM majors (Hyde, Lindberg, Linn, Ellis,
& Williams, 2008). Males still outnumber females among the very high scorers on
mathematics tests (Lubinski & Benbow, 2006), however, the proportion of women who
score very high has increased substantially. For example, in the early 1980s, the ratio of
males to females who score above 700 on the SAT math section was 13:1 (Benbow &
14
Stanley, 1983), but has declined to 3:1 in the past ten years (Halpern et al., 2007).
involvement as the STEM workforce is not populated by only the highest scoring
The gender gap in STEM majors can be explained by factors other than science
and mathematics achievement, but many of these factors are not well understood. One
explanation for the gender gap is that many women enter college with an interest in
STEM, but do not obtain a STEM degree (Huang et al., 2000). Gayles and Ampaw
(2011) found that the following elements influenced degree attainment: high school
science grades, parental education and income, institutional type (e.g., selective and
liberal arts), attending school full-time, relationships with faculty, and CGPA.
Furthermore, they found that there was an interaction between some of these factors and
gender. For example, females were more influenced than males by talking to faculty
occasionally outside the classroom and by attending school full-time (Gayles & Ampaw,
2011). Others found that women are more likely to finish a STEM degree when the
Recent studies have focused on reasons women do not initially select STEM
Some of the reasons discovered include lack of flexible hours and family responsibilities
(Heilbronner, 2013). Shapiro and Sax (2011) also summarized common factors related to
selecting and persisting in a STEM major, such as the competitive culture and pedagogy,
lack of sense of belonging, and family expectations. It should be noted that when women
do major in STEM fields, they perform better than their male counterparts, but face
15
difficulties related to the institutional culture (Huang et al., 2000). Women have also been
found to face difficulties once in STEM occupations. For example, women tend to be
viewed as less competent than men in male-dominated occupations, unless there is clear
evidence in the contrary (Heilman, Wallen, Fuchs, & Tamkins, 2004). Furthermore,
women who are judged to be competent are more likely to be disliked (Heilman et al.,
2004). Both findings may help to explain why women may be less likely to enter the
positively associated with achievement in science (Li, Shavelson, Kupermintz, & Ruiz-
Primo, 2002) and selection of a STEM major (Davison, Jew, & Davenport, 2014);
(Davison et al., 2014). Additionally, when verbal achievement was found to be higher
than math achievement, students were more likely to select non-STEM majors (Davison
et al., 2014). Because females tend to demonstrate higher verbal achievement, Davison et
al. (2014) hypothesized that this finding may partially account for the fact that females
are less likely than males to major in STEM fields, even when they have equal math and
science abilities to males. Lubinski and Benbow (2006) studied the career paths of high
math achievers and found that women were more likely to receive degrees in humanities,
life sciences, and social sciences while men were more likely to receive degrees in
mathematics, computer science, engineering, and physical sciences. They concluded that
these women are working in multidisciplinary fields that utilize both their quantitative
and verbal reasoning abilities, as well as matching their personal preferences (Lubinski
& Benbow, 2006). Therefore, it is possible that it is not necessarily a lack of STEM
16
interest, but rather a stronger interest in non-STEM fields that leads talented females to
These findings highlight the need for further research on the relationship
between gender and STEM. In addition, they emphasize the importance of including
support that benefits both females and males, such as additional STEM experiences
before college, faculty support, support for individuals with family responsibilities, and
the educational timeline and into STEM careers. This project aims to add to the research
STEM education and increase the number of STEM graduates. Many of these initiatives
(CoSTEM, 2013). This section will briefly outline the key initiatives in an effort to
illustrate the recent priority given to STEM in the field of education and to highlight the
The STEM initiatives first began as a result of programs that followed the No
Child Left Behind Act passed by Congress in 2001 (No Child Left Behind [NCLB],
2002). NCLB sought to improve the academic achievement of U.S. students by requiring
states to adopt content standards and to show adequate yearly progress in their students.
Following NCLB, Race to the Top is a competitive grant program and was one of the
the first round of the $4.3 billion dollar competition, the Department of Education gave
17
states preference if they developed strategies to improve achievement in STEM,
established in 2009 (The White House, Office of the Press Secretary, 2009). This
program sought to move students’ science and mathematics achievement from the
middle of the pack to the top of the pack, as compared to other nations. Numerous other
initiatives resulted from this program. For example, in the 2011 State of the Union
Address, the President announced the goal to produce over 100,000 new STEM teachers
over the next decade (The White House, Office of the Press Secretary, 2011). This
announcement led to the 100Kin10 coalition (100,000 teachers in ten years) led by the
Carnegie Corporation of New York which raised over $30 million (CoSTEM, 2013).
Overall, the President’s Educate to Innovate campaign has raised over $1 billion towards
STEM programs (The White House, Office of the Press Secretary, 2015).
In 2010, the White House also began hosting an annual Science Fair in an effort
Science Fair, the President issued a call to action to 200,000 Federal scientists and
engineers to volunteer and try to engage students in STEM (CoSTEM, 2013). At the
2015 White House Science Fair, President Obama announced over $240 million in
STEM commitments (The White House, Office of the Press Secretary, 2015). The goals
of these funds are to inspire and prepare students to succeed in STEM fields, especially
18
In an attempt to increase the quality of STEM education and increase student
created the Federal STEM Education 5-Year Strategic Plan (CoSTEM, 2013). The
STEM Strategic Plan outlined five goals, including 1) improving STEM instruction, 2)
increasing the number of STEM experiences for youth, 3) graduating one million
many of the programs. Both Race to the Top and the STEM Strategic Plan seek to
increase STEM education and career opportunities for women. In addition, the
focusing on gender equity in STEM (e.g., U.S. Department of Commerce, 2011b; U.S.
This section outlined key resources directed towards STEM. These initiatives
increase the number of STEM graduates. Given these goals, it is especially important
that STEM achievement and college readiness in STEM are accurately presented and
interpreted. There is currently very little information directly linking achievement and
college readiness with outcomes for STEM majors. The remaining sections will provide
19
information related to measuring college readiness and modeling achievement growth
in STEM.
College Readiness
recent decades (e.g., NCLB, 2002; U.S. Department of Education, 2010). This is likely
due to several factors. In recent years, the number of students attending college has
approximately 69% enrolled in 2009 (National Center for Education Statistics [NCES],
2011). However, many students are not prepared for college and require remedial classes
(Moore et al., 2010; NCES, 2011). For example, 22%-42% of first-year college students
require remedial education (NCES, 2011). In addition, some students drop out of college
due to academic difficulty and they usually do so in the first year (Choy, 2002). Being
college ready has benefits for both the student and the public (Merisotis & Phipps, 2000);
College readiness can be defined in many ways and definitions include concepts
ranging from academic achievement to leadership and interpersonal skills (e.g., Barnes,
Slate, & Rojas-LeBouef, 2010; Schmitt et al., 2009). Although it is easy to see that a
constrained by the type of data that is readily accessible. Thus, the most commonly used
combination of high school classes, HSGPA, and achievement test scores (Clinedinst,
Hurley, & Hawkins, 2011). For the purpose of this dissertation, college readiness will be
20
Specifically, academic preparedness will be measured by scores on the IAs and academic
readiness. Both the ACT and the SAT have been found to be good predictors of first-year
college grades and degree completion (e.g., Burton & Ramist, 2001; Radunzel & Nobel,
2012; Sawyer, 2010; Willingham, Lewis, Morgan, & Ramist, 1990). In addition, there is
long history of documenting the relationship between the IAs and success in college
(e.g., Ansley & Forsyth, 1983; Loyd, Forsyth, & Hoover, 1980; Qualls & Ansley, 1995;
Wood & Ansley, 2008). These studies used correlations and linear regressions to relate
scores on the IAs to college grades and graduation rates. Recently, Fina (2014) examined
these relationships with an updated sample. This study confirmed previous findings that
higher scores on the IAs were associated with higher first year grade point averages
(FYGPAs). In addition, Fina (2014) set empirical benchmarks on the IAs that link
benchmarks are very useful for quickly conveying information. Most college readiness
courses taken during the first year of college. For example, ACT’s CRBs describe “a
criterion for success for a typical student at a typical college [emphasis in original]”
21
rather than categorized as ready or not. Logistic regression is one way to attach meaning
ACT’s CRBs were calculated using logistic regression and are defined as the
point at which 50% of students receiving that score earned a B or above in the associated
courses (Allen, 2013; Allen & Sconing, 2005). On the IAs, CRBs were established by
Furgol, Fina, and Welch (2011) by linking to ACT scores. In 2014, alternative CRBs
were presented by Fina (2014) that linked the scores on the IAs directly to course grades
for Iowa students using logistic regression. For the remainder of this document, these
CRBs will be referred to as the IAs’ CRBs. They represent the point at which students
receiving the score had a 50% chance of earning a B or above. The IAs’ CRBs were
found to be 336 in Science, 319 in Math, and 317 in Reading (Fina, 2014). They provide
results similar to those seen with ACT’s CRBs (Fina, 2014). In addition, meeting
General CRBs measure students’ ability to succeed in entry level college courses;
however, for reasons outlined throughout this dissertation, general CRBs may not be suitable
indicators of readiness for students planning a STEM major. Camara (2013) echoes this
sentiment when discussing the validity of CRBs and suggests that more rigorous
benchmarks may be needed for STEM majors. Taking into account that STEM is a relatively
new acronym, surprisingly, very little research has specifically examined the college
readiness of STEM students. Researchers at ACT recently began to address this issue.
Mattern et al. (2015) examined over 175,000 students at 27 public four-year institutions and
identified math and science courses that STEM majors were most likely
22
to take in the first year. They found that STEM majors tend to take more math and
science courses in their first year and are more likely to take multiple math and science
courses in the same term. In addition, STEM majors tend to take more advanced math
and science courses than non-STEM majors. See Table 2.1 for a brief summary of their
findings on the most common math and science courses for STEM majors as compared
to non-STEM majors.
Mattern et al. (2015) also established STEM Readiness Benchmarks on the ACT
test. ACT’s STEM Readiness Benchmarks indicate that students require a higher level of
achievement in order to succeed in the first-year math and science courses typically taken
by STEM majors than would be implied if one used the general CRBs. See Table 2.2 for
found that students who met the STEM Readiness Benchmarks were more likely to
persist in a STEM major and to earn a STEM degree. Similarly, the College Board set
benchmarks on the SAT specific to content areas, including Reading, Writing, and STEM
(Wyatt, Remigio, & Camara, 2013). STEM was defined as including course work in
math, science, computer science, and engineering. They found the STEM benchmark to
be higher than their general CRB in Math, 610 as compared to 500 (Wyatt et al, 2013).
Overall, these findings provide further evidence that general CRBs are not the
same as STEM Readiness Benchmarks. STEM Readiness Benchmarks are unique and
valuable measures and offer information not currently available through most
standardized tests. Providing these new benchmarks to students in middle school or high
school would allow for more detailed information on the knowledge and skills necessary
23
interested in pursuing a STEM major, but are not on-track to be college ready for STEM
coursework, these benchmarks would enable them to take preemptive measures. For
guidance counselor could suggest remedial coursework in junior high or high school.
Also, STEM Readiness Benchmarks may pique the interest of students who score well on
the IAs’ Math and Science tests, but who had not previously considered a STEM major.
This dissertation will establish STEM Readiness Benchmarks on the IAs. These
benchmarks are unique to those found by Mattern et al. (2015) as they will be established
on an achievement test taken by all students in the state, as opposed to one developed
mainly for those who plan to attend college. In addition, these benchmarks can be linked
to scores on the IAs in order to provide on-track STEM readiness indicators for students
as young as Grade 6. Finally, these benchmarks will serve to confirm the trends found by
While the Mattern et al. (2015) study was an important first step, there is still a
majors. This void is especially troubling as research suggests that college and career
readiness planning should begin before high school (ACT, 2008). To address this void,
this dissertation also seeks to model achievement growth on the IAs for STEM majors.
readiness is determined by educational experiences that occur in the years prior. In fact,
readiness than achievement changes that occur during high school (ACT, 2008).
24
Therefore, college and career readiness planning should begin before high school. This
section will begin with a brief review of research on early achievement differences in
science and math and will follow with research on latent growth models. It will then
As previously discussed, there are often small gender differences in science and
math on high school tests (e.g., Lubinski & Benbow, 2006; Quinn & Cooc, 2015). In an
attempt to better understand when these differences develop, researchers have examined
gender differences in early grades. One study found that gender differences in science
was small in early grades and became more pronounced through high school (Jones,
Mullis, Raizen, Weiss, & Weston, 1992). However, more recent studies have found other
trends. For example, in a longitudinal study on Grades K-8, kindergarten students did not
show gender gaps in math; however, differences did develop in elementary school, but
the gap then began to narrow in middle school (Robinson & Lubienski, 2011). Quinn and
Cooc (2015) also found a gender gap in science achievement in Grade 3 that narrowed
slightly by Grade 8.
The IAs not only have a long history of documenting relationships between
achievement scores and college readiness, but also in measuring student growth. The IAs
discussed in Fina (2014), the IAs are uniquely suited for measuring growth and college
readiness in a way that most other states’ achievement tests are not. Specifically, they are
designed to capture more permanent aspects of learning (Lindquist, 1942), beyond the
towards college readiness, Fina (2014) found that the IAs are useful for making on-track
25
interpretations about college readiness. In addition, different developmental trajectories
were found to be associated with different levels of performance in college. The analyses
growth, such as that achieved through LGMs. There are many definitions of growth
(Castellano & Ho, 2013), but this dissertation focuses on growth as measured by LGMs.
Researchers have examined a variety of factors related to growth; however, there have
been limited studies on achievement in science and math and no studies on STEM
majors. Therefore, this section will discuss LGM findings, although most are not specific
LGMs have been used to identify whether variations in growth are better
explained at the school level or individual level (Muthén, 1997; Williamson, Appelbaun,
& Epanchin, 1991). Others have examined adding covariates to LGMs, such as gender or
English Language Learner (ELL) status. Wang, Chen, and Welch (2012) examined
growth trajectories for Hispanic students and ELL students and found they had similar
patterns of growth for ACT takers; however, there were differences in the growth patterns
of students who did not take the ACT. Interestingly, Muthén and Khoo (1998) found that
even when there is not a significant difference in achievement scores by gender, LGMs
can still identify differences in growth rates for males and females. The alternate is also
true, even when there are differences in achievement scores, LGMs may not find
differences in growth rates (Zvoch & Stevens, 2006). LGMs have also been used to
predict performance. For example, Choi and Goldschmidt (2012) used LGMs to predict
performance on the California High School Exit Exam. They found that Grade 9 was a
26
better predictor than achievement growth in high school. Muller, Stage, and Kinzie
(2011) examined science achievement growth trajectories with gender and race-ethnicity
as covariates. They found substantial differences in the growth trajectories for females,
African American students, and Latina students. For these groups, they found that there
were differences not only at Grade 8, but also in growth rates in middle school and high
school. They concluded by calling for further research that disaggregates growth data by
with less STEM attrition and greater success as a STEM major (e.g., LeBeau et al., 2012;
Westrick, 2015). In addition, achievement growth before high school has been shown to
students, educators, and administrators. For example, it will provide information on the
typical levels of achievement of STEM majors when they were in Grade 6 or Grade 8. It
may be that STEM majors have different rates of growth than non-STEM majors.
Alternately, it may be that STEM majors have the same rates of growth as non-STEM
majors, but start with higher early achievement levels. These findings would allow for
early remediation if students were not on track for STEM readiness, but were interested
in majoring in STEM. In addition, it may help to identify students that show promise for
success in STEM fields. Similarly, measuring growth of STEM majors by including the
covariate of gender may also help to identify any differences between females and males
27
In order to address the lack of research linking grade school achievement to
college outcomes of STEM majors, this dissertation will compare the growth of STEM
majors and non-STEM majors on the IAs starting in Grade 6. It is hypothesized that
STEM majors will start at a higher level of achievement, but it is unknown whether the
difference between STEM majors or non-STEM majors would remain constant over time
or possibly increase. In addition, gender differences for these two groups will also be
examined. Growth will be compared through latent growth models (LGMs). The
following section will describe LGMs and typical procedures and steps taken during the
different achievement levels and tend to grow at different rates. LGMs take into account
both of these factors and offer a flexible way to model students’ growth trajectories.
LGMs are commonly approached from one of two perspectives, including the multilevel
model (MLM) and the structural equation model (SEM). These two approaches give
identical estimates when equivalent models are constructed (Hox & Stoel, 2005). This
regression model with a mean structure (Kline, 2011). As outlined in Kline (2011), an
LGM in SEM requires that measures are repeated on at least three occasions and that the
data are time structured. When the LGM is constructed as an SEM, the measurement
model can be specified as follows (Preacher, Wichman, MacCallum, & Briggs, 2008):
y = τ + Λη + ε. [2.1]
28
In this equation, y is the p-dimensional vector of the repeated measure for an
of disturbance terms.
From this model, the covariance structure (2.2) and mean structure (2.3) can be
Σ = ΛΨΛ′ + Θε [2.2]
μ = τ + Λα. [2.3]
m matrix of factor loadings, and α is a p-dimensional vector of the latent variable means.
interpretation of the resulting means. In addition, loadings on the intercept factor are
fixed to 1.0 and the loadings on the slope factor are fixed to represent growth over time.
Growth is most often defined linearly, but can be defined non-linearly, such as by the
addition of a quadratic term. In addition, while growth is most often modeled as a single
overall linear trajectory, it can also be modeled piecewise to account for different phases
that arise due to special circumstances, such as standardized score changes on tests or an
intervention that occurs partway through the repeated measures. When growth is
modeled piecewise, it can be specified as linear or non-linear for each individual piece
(Kohli, Harring, & Hancock, 2013). For example, if growth were expected to be linear
before an intervention, but the trajectory has curvature after the intervention, it may best
29
be described by a piecewise liner-quadratic LGM. Alternately, if growth were expected
to be linear both before and after a score change, a piecewise linear-linear LGM could be
specified. Piecewise growth models offer flexibility and are more appropriate when
investigating growth that may have specific phases (Chou, Yang, Pentz, & Hser, 2004;
Growth modeling follows an iterative process. This process usually begins with
simple models and progresses to more complex models. Substantive theory and practical
considerations guide the growth modeling process. Kline (2011) suggests that LGMs be
analyzed in two steps. The first step uses latent variables to model the growth of the
repeated measures. If the resulting unconditional model has adequate fit, then covariates
may be added to the model to help better explain change over time. Separating the
process into these steps allows one to more easily identify sources of poor model fit.
Summary
underrepresented in STEM fields and the reasons for this gender disparity are not well
understood. Increasing women’s participation in STEM is one way educators are hoping
to meet the goal of 1 million additional STEM graduates in the next ten years (CoSTEM,
achievement growth patterns of women that may help to explain why women are less
likely to pursue STEM degrees. Reducing STEM attrition is another way to achieve the
goal (PCAST, 2012). STEM success has been shown to be associated with higher levels
of high school academic achievement (e.g., Mattern et al., 2015); however, the level of
30
academic achievement needed is unknown. As students are encouraged to consider
STEM majors, it is especially important to ensure that they are adequately counseled on
the level of academic achievement needed to succeed in STEM fields. This dissertation
STEM through establishing STEM Readiness Benchmarks and modeling growth for
STEM majors.
31
Table 2.1
Summary of the Distribution of Students’ First College Math and Science Courses
by STEM Category
Table 2.2
Comparison of ACT’s College Readiness Benchmarks
and STEM Readiness Benchmarks
32
CHAPTER III: METHODS
This study proposes to investigate the achievement growth and college readiness
of STEM majors. Chapter III details the data and methods that will be used to address
the three objectives outlined in Chapter I. To begin, a brief background on this study is
provided and the data sources are described. Next, latent growth models (LGMs) are
outlined for comparing the achievement growth of STEM majors and non-STEM majors.
In addition, the methods for calculating and validating the STEM Readiness Benchmarks
Background
This dissertation utilizes the same data sources as Fina (2014) and the matching
procedures are similar to those used in Fina (2014). Differences between this study and
Fina (2014) arise in the research questions and analyses. Fina (2014) measured the
achievement growth of all students, regardless of college status, and general college
readiness. As a result, the datasets included students who did not attend college, as well
as students who enrolled in first-year general education college courses. In contrast, this
dissertation focuses on the achievement growth and college readiness of STEM majors.
Therefore, the final datasets are a subset of those used in Fina (2014). Specifically, the
two types of datasets in this study include 1) students who attended college and 2)
students who enrolled in first-year STEM courses. Similarities and differences between
the two studies will be discussed throughout the Methods and Results sections.
33
Data Sources
For this dissertation, data came from two sources.1 The sources consisted of 1)
Iowa Testing Programs (ITP) Archival Data and 2) the University of Iowa Registrar
Records. The data span the 2001-2002 to 2012-2013 academic years. From these
sources, several datasets were created. These sources are identical to those utilized by
Fina (2014) and the matching procedures are similar; however, the subsequent datasets
are subsets of those in Fina (2014). Table 3.1 contains a summary of each source and the
The state achievement tests consist of subtests of the ITBS (Hoover et al., 2001)
and the ITED (Forsyth et al., 2001), which will collectively be referred to as the Iowas
(IAs). The IAs are a battery of large scale achievement tests designed to assess student’s
strengths and weaknesses in different subject areas. In addition, they have been
The achievement records on the IAs for the high school graduating class of 2008
were matched from middle school to high school. This cohort was selected because it
allows for the availability of the subsequent five years of college data. The data
included achievement scores and demographic information. For this dissertation, scores
from Grades 6-11 were included for Science (SC), Math Total (MT), and Reading
Comprehension (RC). Scores included both the National Standard Score (NSS) and the
National Percentile Rank (NPR). Students were included in each sample if they had an
1
Prior to receiving the University data, the Internal Review Board confirmed its exempt status. All
individuals with access to the data are FERPA certified.
34
NSS for Science, Math, or Reading in Grade 11 and at least two years of data in Grades
6, 7, and 8.
continuum. The NSS scale is a vertical scale and average NSSs increase with grade level,
readily allowing for monitoring growth (Forsyth, Ansley, Feldt, & Alnot, 2003; Hoover,
Dunbar & Frisbie, 2003). They are scaled such that 227 is the median performance of
students in the spring for Grade 6 and 275 for Grade 11. See Table 3.2 for a summary of
the median NSSs for Grades 6-11. The scale reflects empirically established decreasing
gains and increasing variance in scores as students advance in school. The NSS scale is
not affected by time of year, but the NPR scale is dependent on when students test. NPR
nation who were in the same grade and tested in the same norming period (Forsyth et al.,
2003; Hoover et al., 2003). There are three norming periods for the NPR scale: Fall,
Midyear, and Spring. See Table 3.3 for a summary of the test dates that coincide with
each norming period. For this dissertation, the NSSs will be adjusted to the Fall norming
period for all of the analyses (Forsyth et al., 2003; Hoover et al., 2003). Further details
information with information collected by the University of Iowa Admissions Office for
Iowa students who applied to and attended the University in the 2008-2009 school year.
The transcript data comprised college courses taken, course grades, credit hours earned,
first-year grade point average (FYGPA), and four-year cumulative college grade point
35
average (CGPA). In addition, college enrollment, college major, and graduation status
were included. The admissions data included students’ high school grade point averages
For the 2008-2009 school year, 15,582 first-time, first-year students applied and
12,827 were admitted (University of Iowa, 2009). Policies prior to 2009 dictated that
students were automatically admitted if they took the required high school courses and
ranked in the top 50% of their graduating class (Board of Regents, 2009). Of the
admitted students, 4,182 enrolled and 54% of these were in-state (University of Iowa,
2009). The one-year retention rate was 83% (Board of Regents, 2013). The resulting
dataset for this dissertation included approximately 2,000 in-state college students who
Matching records from the two sources was done in two steps, as described in
Fina (2014). In the first step, student records from the ITP archival data for Grades 6-11
were matched across grades. Matching was completed in a stepwise fashion beginning
with Grade 11. A SAS computer program was utilized to match records based on student
State ID numbers (SAS, 2011). When students were missing State IDs, they were
matched using exact matches on first name, last name, school district code, and birthdate.
Finally, remaining students were matched using fuzzy matches and then visually
inspected to ensure the accuracy. Fuzzy matches denote imperfect matches, such as a
student with a correct birth month, but not matching birth year, or situations in which
nicknames may have been used. Records that were unable to be matched were excluded
36
The next step required first matching records from the University of Iowa
Admissions Office and Registrar’s Office on University Student ID. This was necessary
because the data were provided in a relational database, but the analyses required that
each record contain all the relevant information for that student. The resulting dataset
was then matched to the ITP Archival data using fuzzy matching.
Study Procedures
The purpose of this study is to compare the growth and college readiness of
STEM majors versus non-STEM majors. In order to accomplish this, the three research
37
3. Examine and evaluate the validity evidence for the STEM Readiness
b. Calculate the error rates and PCC for selected individual Science and
Math courses. Compare these rates with those found using all courses
differently by course.
c. Calculate the error rates and PCC for the STEM Readiness
major), and growth from Grades 6 to 11. Compare these rates through
STEM majors and non-STEM majors. First, characteristics of STEM majors and non-
STEM majors will be summarized. Next, latent growth models (LGMs) comparing
achievement growth in Science, Math, and Reading will be developed and evaluated to
38
For these analyses, datasets were created that matched students’ records on the
IAs to their majors at graduation. See Appendix A for a list of identified STEM majors
and their required introductory science and math courses. Three datasets were created to
represent the three subject areas: Science, Math, and Reading. In order to be included in
the datasets for Objective 1, students were required to have a Grade 11 NSS on the IAs in
the subject of interest (i.e., Science, Math, or Reading) and to have enrolled at the
University of Iowa in 2008. In addition, students were required to have scores for at least
2 of the 3 years in Grades 6, 7, and 8 in the subject of interest. Because districts may take
the statewide achievement test during any of three norming periods (Fall, Midyear,
Spring), the NSSs were adjusted to the Fall norming period for all students to allow for
STEM majors and non-STEM majors. In addition, means and standard deviations will be
provided for HSGPA, FYGPA, CGPA, and scores on the IAs for Science, Math, and
descriptives will also be provided by gender for STEM majors and non-STEM majors.
Latent growth models. Growth modeling typically begins with simple models
and progresses to more complex models. Following this process, models of increasing
complexity will be examined until unconditional models are found for each subject area
that achieve adequate fit, then STEM status and gender will be added as covariates. If a
linear model seems reasonable based on the data, then simple linear LGMs will be
39
Fina (2014) modeled the growth of all Iowa students, regardless of whether they
attended college, starting in Grade 6. In that study, a simple linear LGM was examined,
then a quadratic term was added, but neither of these models demonstrated adequate fit.
For the final LGM, the Grade 11 time score for the slope parameter was allowed to be
estimated from the data. This model was labeled LGM-Free (Fina, 2014). For future
Because this study only utilizes college students when examining the growth of
STEM majors as compared to non-STEM majors, one would expect different growth
trajectories than those that were estimated in Fina (2014). In addition, one would expect
a higher intercept mean in Grade 6 for the sample of college students versus the full
population of students. Therefore, this study will not utilize the LGM parameters
established by Fina (2014). However, this study will follow similar procedures to those
outlined by Fina (2014). In accordance with Fina (2014), this dissertation will use the
following procedures to establish the LGMs for Science, Math, and Reading.
Linear and quadratic LGMs. To begin, graphs of the observed mean NSSs by
grade level for each subject will be examined. If a linear model seems reasonable based
on the data, then a simple linear LGM will be fit to the data, specifically a structural
regression model with a mean structure. Growth is represented by two latent variables:
initial status, or IW, and linear change, or SW. IW represents the intercept, and the
loadings from the repeated measures to IW are fixed at 1 for all time points for
achievement in the subject of interest in Grade 6, corrected for measurement error. The
40
variance of IW represents the variability in the individual level intercepts around the
mean of IW. In contrast, SW represents the change over time. For a linear model, the
loadings are fixed to constants that represent equidistant time points (0, 1, 2, 5). The
mean of SW represents the average linear trajectory across all students in science
growth.
If the graphs of the observed mean NSSs indicate that student growth is not linear,
but rather quadratic in shape, then a quadratic slope factor, QW, can be added. With a
quadratic term, the additional factor would represent curvilinear growth (with loadings 0,
1, 4, 25). For both the linear and quadratic models, the intercept factor and slope factor
will be allowed to covary. In addition, the residual variances of the outcome variables
will be allowed to vary across time and will not be correlated, in line with conventional
Piecewise LGMs. Additionally, a piecewise LGM with a mean structure can be fit to
the data. The basis for this decision would come from examining the observed means and the
fit of the previous LGMs. This model is very similar to the linear LGM. Growth will still be
represented by two latent factors, IW and SW. The difference arises in the specification of
the mean structure. While the IW loadings will still be fixed to 1 for all the time points
(Grades 6-11), the regression intercept will not be fixed at zero at Grade
Model evaluation. All LGM analyses will be conducted in Mplus (Muthén &
Muthén, 1998-2012). Each model will be presented separately. To begin evaluating each
41
model, the signs and magnitudes of the parameter estimates and standard errors will be
examined to ensure that they are reasonable and in-line with expectations (Kline, 2011).
Next, model fit will be assessed using the following indicators: Bayesian Information
Criteria (BIC; Schwarz, 1978), Root Mean Square Error of Approximation (RMSEA),
Comparative Fit Index (CFI; Bentler, 1990), and Standardized Root Mean Square
Residual (SRMR). It is recommended to use multiple fit indexes, as some indices are
more robust than others (DeRoche, 2009). Each of the indicators examines a different
aspect of model fit and there are recommended guidelines for assessing the resulting
guidelines will be described in the Results section. If any of the models for Science,
Math, or Reading do not show adequate fit, the identified model(s) will be respecified,
Once each model adequately described the data, the covariates of STEM status,
gender, and the interaction between STEM status and gender will be added. See Figure
3.1 for a graphical representation of a conditional linear LGM with the covariates. See
Figure 3.2 for an example of a conditional LGM with a quadratic term. Modeling the
individual-level predictors in this way allows one to examine the direct effects of the
covariate on the slope and intercept (Stoel, van den Wittenboer, & Hox, 2004). For
STEM status, STEM graduates who completed the degree within five years will be coded
1 and all other college students will be coded 0, regardless of graduation status. For
gender, males will be coded 0 and females will be coded 1. Significance of the covariates
will be checked. If significant at the .05 level, then the magnitude of the effects on slope
42
Objective 2: Establishing STEM Readiness Benchmarks on the IAs
benchmarks will serve as indicators of science and math readiness for students planning
to major in a STEM field. For these analyses, datasets were created that matched
students’ NSSs in Grade 11 with their course grades from typical first-year STEM
courses at the University of Iowa. See Appendix B for a list of typical first-year STEM
courses. While Fina (2014) did not adjust NSSs when setting the general CRBs, NSSs
were adjusted to the Fall norming period when establishing the STEM Readiness
Benchmarks. For Fina (2014), the general CRBs represent a level of achievement needed
for college readiness that can be applied across all norming periods in Grade 11;
however, for this study, the STEM Readiness Benchmarks represent a level of
achievement needed for college readiness in STEM courses in Fall of Grade 11.
Typical first-year STEM courses were identified by Mattern et al. (2015) when
establishing the STEM Readiness Benchmarks on the ACT. In order to confirm that these
courses would also be appropriate for this sample, required courses for STEM majors and
typical first-year math and science courses taken by students who graduated from the
University of Iowa with STEM majors were examined.2 The courses were found to be
equivalent to those identified by Mattern et al. (2015). For Math, benchmarks will be set
based on performance in Calculus classes. For Science, benchmarks will be set based on
Engineering. Also consistent with Mattern et al. (2015), benchmarks will be established
2
Additional details on identifying typical first-year STEM courses are provided in Chapter IV.
43
typical first-year STEM courses will be included in the analyses. Because students often
take multiple science and math courses the first year of college, students may be
All benchmark analyses will be conducted in Mplus (Muthén & Muthén, 1998-
2012) using logistic regression models to link Grade 11 performance on the IAs in
Science and Math to course grades. The STEM Readiness Benchmarks represent the
level of performance on the IAs that is associated with a .50 probability of earning a B or
above in typical first-year STEM courses. This is the same level of success utilized in
setting the benchmarks in both Mattern et al. (2015) and Fina (2014).
particular grade in science or math courses is associated with students’ scores on the IAs
obtaining a B or above given students’ scores on the IAs. For the benchmark analyses,
students’ grades will be coded such that a grade of B or above is 1 and a grade below a B
is 0 (i.e., Stem Ready (SR) = 1 or SR = 0). The regression model may be specified as
The probability of the obtaining a B or above given a particular NSS may be expressed as
The logits (logistic regression coefficients) and thresholds will be examined and
summarized. If the estimates are significant at the .05 level, then the benchmark will be
44
Objective 3: Provide Validity Evidence for the STEM Readiness Benchmarks
The Standards offer guidelines for sound testing practices and for assessing the validity
APA, & NCME], 2014). Specifically, The Standards (AERA, APA, & NCME, 2014)
call for validity evidence for all proposed score interpretations and intended uses of
test scores, and for evidence on the relationship between test performance and the
relevant criteria when establishing cut scores or benchmarks (e.g., 1.2, 5.23, 12.2).
indicators in STEM courses, false-positive error rates, false-negative error rates, and the
percentage of correct classifications (PCC) for the STEM Readiness Benchmarks will
be calculated. These rates will then be compared with those found using the general
CRBs developed for typical first-year general education courses by Fina (2014). It is
expected that the STEM Readiness Benchmarks will have higher PCCs and lower false-
positive error rates than the general CRBs when applied to STEM coursework, because
of differences in difficulty levels for typical first-year STEM courses versus typical
first-year general courses. However, because it is expected that the STEM Readiness
Benchmarks will be higher than the general CRBs, they may also have higher false-
Next, the error rates and PCC will be calculated for selected individual Science
and Math courses. These rates will be compared with those found using all courses to
45
Finally, error rates and PCC will be calculated for the STEM Readiness Benchmarks by
group, including by STEM status, gender, and achievement growth on the IAs from
Grades 6 to 11. Chi-square goodness of fit tests will be run to determine if there were
differences in the error rates and PCC when examining the STEM Readiness
Benchmarks by STEM status, gender, and growth from Grades 6 to 11. These results will
indicate whether students who met the benchmarks were more likely to earn a STEM
degree. They will also indicate whether the STEM Readiness Benchmarks function
IAs from Grades 6 to 11. Descriptive statistics based on student’s benchmark status,
STEM status, gender, and achievement growth from Grades 6 to 11 will be reported.
Summary
field of education today (CoSTEM, 2013), there is a lack of research examining how
student achievement growth and college readiness vary for STEM and non-STEM
majors. This dissertation seeks to address these gaps. Objective 1 will compare the
achievement growth of STEM and non-STEM majors, from middle school to high
school, through latent growth modeling. In addition, it will examine whether there are
Objective 2 will establish STEM Readiness Benchmarks on the IAs for STEM majors’
typical first-year science and math courses taken at the University of Iowa. The STEM
46
student’s readiness for a STEM major than general CRBs (Mattern et al., 2015).
Furthermore, they may allow for remediation, if necessary, or may help to encourage
interest in STEM fields for students who are accademically gifted in science and math.
Objective 3 will establish validity evidence for the STEM Readiness Benchmarks. The
findings from this dissertation will provide invaluable information to students, STEM
47
Table 3.1
Summary of Data Sources
48
Table 3.2
Median National Standard Scores (NSS)
Grade 6 7 8 9 10 11
Table 3.3
Norming Periods by Test Dates
49
Figure 3.1 Conditional linear latent growth model.
50
Figure 3.2 Conditional latent growth model with a quadratic term.
51
CHAPTER IV: RESULTS
Chapter III outlined the data and methods used in this study. Chapter IV reports
the results of the study. Results are organized by the three objectives outlined in Chapter
III. First, the latent growth model (LGM) results are presented. Specifically, similarities
and differences in achievement growth in Science, Math, and Reading for STEM majors
and non-STEM majors are described and gender differences are highlighted. Next, the
STEM Readiness Benchmarks in Science and Math are calculated and compared to
general college readiness benchmarks. Finally, validity evidence for the STEM
For the remainder of this dissertation, STEM major and STEM graduate will be
used interchangeably to indicate a student who graduated with a STEM major. Of the
non-STEM majors, 95% graduated and the remaining 5% either dropped out of college
graduates and students who may not have graduated, because the purpose of this study
was to compare STEM graduates with those who did not graduate with a STEM degree,
whether that was because they graduated with a non-STEM degree or because they left
the university.
Science, Math, and Reading for STEM majors and non-STEM majors. As outlined in
Chapter III, analyses will begin with simple linear LGMs and move to more complex
models, as needed. While hypothesis testing is often the focus of LGMs, it will not be for
52
this study. Rather, the LGMs will be used for descriptive purposes. Specifically, the
LGMs will describe the growth of students in Science, Math, and Reading, and then
compare the growth of STEM majors and non-STEM majors. Summaries of the
unconditional models will be given first and then covariates will be added and discussed.
To begin the process, descriptive statistics for the LGM dataset were examined, including
the mean scores on the Iowas (IAs) for each subject at Grades 6, 7, 8, and 11.
Descriptive Statistics
National Standard Score (NSS) on the IAs in the subject of interest (i.e., Science, Math,
or Reading), to have enrolled at the University of Iowa in the 2008-2009 school year, and
to have scores for at least 2 of the 3 years in Grades 6, 7, and 8 in the subject of interest.
The resulting datasets for Science, Math, and Reading were very similar, as most
students took each test every year. In total, Reading had 1346 students, Math had 1344
students, and Science had 1327 students. Because they are so similar, general descriptive
statistics will be reported first for the group as a whole, based on the largest dataset
(Reading). Additional descriptive statistics will be reported later by subject area as each
In total, there were slightly more females than males (53% and 47%,
respectively). Of the 1,346 students, 144 were STEM graduates (11%) and 1,202 were
non-STEM majors (89%). Twenty-nine percent of the STEM graduates were female and
71% of the STEM graduates were male. Overall, students averaged almost a 3.6 GPA in
high school and just over a 3.0 GPA for both the first year of college and for college as a
whole. Females had higher average GPAs, as did STEM graduates. This same trend was
53
seen in female STEM graduates versus male STEM graduates, except that both groups
had the same final college GPA. See Table 4.1 for a breakdown of GPAs for high school
(HSGPA), first year of college (FYGPA), and final college (CGPA) by STEM major and
gender. Students were high achieving, as compared to students in the nation, based on
National Percentile Ranks (NPRs) on the IAs in Science, Math, and Reading. For
example, Grade 11 NPRs were 87, 86, and 82 in Science, Math, and Reading,
respectively; whereas Grade 11 NPRs have been normed to be 50 at the national level for
each subject. See Table 4.2 for mean NSSs and NPRs in Science, Math, and Reading on
the IAs for Grades 6-11. This sample was high achieving not only as compared to
students in the nation, but also when compared to the population of Iowa students
described by Fina (2014), which includes students who did not attend college. This is as
expected, since this sample only includes students who attended college. Furthermore, the
whole, based on Grade 11 NPRs (86, 85, and 81 in Science, Math, and Reading for non-
STEM majors). However, the non-STEM majors score below the STEM majors as the
Grade 11 NPRs for STEM majors were 94, 95, and 88 in Science, Math, and Reading,
respectively. See Appendix C for mean NSSs and NPRs on the IAs disaggregated by
The LGMs were used to help describe the growth of students in a clear and
straightforward manner, and also to allow for the examination of the effects of covariates
in the model. To begin, graphs of the observed mean NSSs by grade level for each
subject were examined. If a linear model seemed reasonable based on the data, then
54
simple linear LGMs were examined to see if they did adequately represent the data. If
the simple linear LGM had adequate fit and reproduced the observed means reasonably
well, then covariates of STEM and gender were added to the model. When a simple
linear LGM did not have adequate fit based on fit statistics or the ability to reproduce
the observed means, graphs of the means were reexamined and a new LGM was
proposed. Once an unconditional LGM was found that accurately described the data,
Science
For Science, the observed mean NSSs for Grades 6, 7, 8, and 11 were examined.
Based on the data, a simple linear LGM was examined first. This model estimated 9
parameters, including six variances (two for the latent factors and four measurement
errors, one for each measurement occasion), one covariance, and two latent factor
The estimates of this model, along with estimates of all models for Math and
Reading, can be found in Table 4.3. This LGM describes a sample of individual growth
trajectories in which the mean intercept (IW) is 263.02 with a standard deviation of
22.21. According to the model, the standard deviation indicated that approximately 68%
of the students start within 22.21 points of the mean of 263.02 on the NSS scale. The
mean intercept is a measure of the average initial status or initial achievement level in
Grade 6. The mean slope (SW) is a measure of the average slope of the individual growth
trajectories. The mean slope for this LGM is 13.77. These results indicate that, on
average, student’s yearly achievement growth is 13.77 points on the NSS scale. At the
55
NSS scale from Grade 6 to Grade 11 (Forsyth et al., 2003; Hoover et al., 2003).
Therefore, this group is showing more growth, on average, than students in the nation.
The variance was significant for IW indicating that there is significant variability
in the individual starting achievement levels in Grade 6. However, the variance was not
significant for SW, indicating that there is not significant variability in the slopes of
individual growth rates from Grades 6-11. The model estimated correlation between IW
and SW was -0.24 indicating that there is a weak negative relationship between a
students were taken from the population and their individual growth trajectories were
plotted, along with the population mean (Figure 4.1). As expected based on the results
above, the majority of the students start within 22 points of IW and there is not a strong
relationship between where a student starts and her or his growth trajectory. There is,
tend to not have a straight growth trajectory, but show a lot of variation in their individual
growth year to year, even though as a group they do not show significant variability in
SW from Grades 6-11. It should be noted that although 30 is a relatively small random
ensure that this first random sample of 30 students was representative of the group as a
whole, many random samples of 30 were generated and plots were inspected. For
consistency, this same random sample will be used when creating the plots for Math and
Reading.
56
The next step was to examine model fit. It is recommended to use multiple fit
indexes, as some indices are more robust than others (DeRoche, 2009). Furthermore,
assessment of fit is not an exact science, but rather the consideration of many factors.
Each of the indicators examines a different aspect of model fit and there are
recommended guidelines for assessing the resulting values of the indicators. The
Bayesian Information Criteria (BIC) is a comparative index and a smaller BIC indicates
better model fit (Nylund, Asparouhov, & Muthén, 2007). The BIC will be used to help
assess whether there is an improvement in model fit when more than one model is
examined. The Comparative Fit Index (CFI) assesses the relative improvement in fit of
the experimental model as compared to the baseline LGM. Values greater than 0.95
indicate adequate fit (Hu & Bentler, 1999). The CFI was found to be the more robust
measure when compared in a Monte Carlo simulation with several other measures,
including the Root Mean Square Error of Approximation (RMSEA; DeRoche, 2009).
Therefore, if there is not a consensus of adequate fit based on multiple indicators, results
from the CFI will be more heavily weighed. The RMSEA index ranges from 0 to 1, with
zero indicating best fit. There is a lot of debate surrounding the cutoff for RMSEA. In
general, a 90% confidence interval containing the value 0.06 or less is desired (Hu &
Bentler, 1999). A 0.07 as the upper limit for the value of RMSEA has also been proposed
(Steiger, 2007). The Standardized Root Mean Square Residual (SRMR) is a measure of
how closely the estimated correlation matrices match the observed correlation matrices.
In general, values of 0.08 or smaller indicate adequate fit (Hu & Bentler, 1999).
57
Overall, this model was found to have adequate fit. See Table 4.4 for a summary
of the model fit statistics for all LGMs. For the simple linear LGM, the CFI at 0.985 was
above the recommended 0.95. In addition, the RMSEA at 0.069 was below the upper
limit of 0.07. The SRMR was 0.098 and while this is above the recommended cutoff of
0.08, the SRMR is higher when sample sizes are large (DeRoche, 2009). Finally, a
comparison of the observed and estimated means shows that the model does reproduce
the observed means well (Table 4.5). To get the estimated means for each grade, the
mean of the latent intercept factor is added to the mean of the latent growth factor and
multiplied by time. For example, Grade 6 (time 0) = 263.02 (1) + 13.77 (0) = 263.02 and
Grade 7 (time 1) = 263.02 (1) + 13.77 (1) = 276.79. Now that an unconditional model
has been found that adequately represents the data, a conditional model can now be
examined. Results from the conditional model will be discussed in the covariate section,
following the results of the unconditional models for Math and Reading.
Math
For Math, the observed mean NSSs for Grades 6, 7, 8, and 11 were examined.
Based on the data, a simple linear LGM seemed plausible and was examined first. The
estimates of this model can be found in Table 4.2; however, this model was not found to
have adequate fit for any of the fit statistics (see Table 4.3). In addition, a comparison of
the observed and estimated means shows that the model does not reproduce the observed
means well (Table 4.5). In particular, it is overestimating Grade 6 and Grade 11. In an
effort to determine the cause of poor fit, the graph of growth trajectories were examined
again and the disturbance terms were considered. As illustrated in the graph (Figure 4.2),
the slope seems to flatten somewhat between Grade 8 and Grade 11. In Table 4.3, Grade
58
11 has the largest standardized disturbance at 0.35. In light of these findings, another
LGM was examined in which the data were not forced to be linear, but rather, the Grade
11 time score for SW was allowed to be estimated from the data.1 The model is labeled
The Free LGM describes a sample of individual growth trajectories in which the
mean intercept (IW) is 253.33 with a standard deviation of 19.99. According to the
model, the standard deviation indicates that approximately 68% of the students start
within 19.99 points of 253.33 on the NSS scale. The mean slope (SW) for this LGM is
16.87 with a standard deviation of 2.09. These results indicate that, on average, student’s
yearly achievement growth is 16.87 points on the NSS scale and, according to the model,
approximately 68% of the students fall within 2.09 points of 16.87 each year. Overall,
students have a lot of variation in their starting achievement levels in Math in Grade 6
and are also demonstrating more yearly growth, on average, than students in the nation.
In addition the variances for the model were significant for both IW and SW, indicating
that there is significant variability in both the starting achievement level at Grade 6 and in
For the Free LGM, the estimate for the Grade 11 time score is now 4.02, instead of
the original value which was set at 5 and based on equidistance in time from Grade 6 to
Grade 11. This estimate for Grade 11 suggests that student growth achievement decelerates
on the NSS scale between Grades 8-11 as compared to the rate of student growth
achievement from Grades 6-8. The model estimated correlation between the intercept
factor and slope factor was -0.17, indicating that there is a very weak negative
1
As recommended by Fina (2014), a piecewise LGM was also examined, but did not improve model
fit over the liner LGM.
59
relationship between a student’s initial status and that student’s rate of growth. The
Model fit was then examined. Overall, the Free model was found to be an
improvement over the linear model and to have adequate fit (see Table 4.4). The BIC
index improved by almost 300 points from a 44092.15 for the Linear LGM to a 43806.30
for the Free LGM. A change in the BIC index over 10 indicates very strong evidence for
preferring the model with the lower BIC value (Kass & Raftery, 1995). The RMSEA and
SRMR were higher than the recommended cutoffs, but the CFI was at 0.988. In addition,
a comparison of the observed and expected means shows that the model reproduced the
observed means well (Table 4.4). Covariates will be added to this model and the results
Reading
Based on the observed mean NSSs in Reading for Grades 6, 7, 8, and 11, a simple
linear LGM seemed plausible; however, the model estimation did not terminate normally.
Therefore, the correlation between IW and SW was set to zero; this allowed the model
estimation to proceed. The estimates of this model can be found in Table 4.3. This model
was not found to have adequate fit for most of the fit statistics (see Table 4.4), however, a
comparison of the observed and expected means shows that the model does not
reproduce the observed means well, especially for Grade 7 (Table 4.5). In an effort to
determine the cause of poor fit, the graph of growth trajectories were examined and the
disturbance terms were considered. Examination of the graph finds an increase in the
60
slope of achievement growth from Grade 6 to Grade 7 as compared to the rest of the
graph (Figure 4.3). For the disturbance terms, Grade 11 has the largest standardized
disturbance at 0.45, followed by Grade 7 at 0.28 (Table 4.3). While Grade 11 showed the
larger disturbance term, the model had the hardest time estimating the mean for Grade 7.
This issue with Grade 7 appears to be the reason the model did not converge initially.
Therefore, a new LGM was examined in which the data were not forced to be linear, but
rather, the time score from SW for Grade 7 was allowed to be estimated from the data. In
this model, IW and SW were allowed to correlate.2 The model is labeled Free in Tables
4.3-4.5.
The Free LGM describes a sample of individual growth trajectories in which the
mean intercept (IW) is 254.28 with a standard deviation of 23.03. According to the model,
the standard deviation indicates that approximately 68% of the students start with a mean
within 23.03 points of 254.28 on the NSS scale. The mean slope (SW) is 13.61 and the
variance of the slope was not significant. These results indicate that, on average, student’s
yearly achievement growth is 13.61 points on the NSS scale. Overall, students have a lot of
variation in their starting achievement levels in Reading in Grade 6 and are also
demonstrating more yearly growth, on average, than students in the nation. In addition, the
variance was significant for IW, but not SW, indicating that there is significant variability in
the individual starting achievement levels at Grade 6, but not in individual growth rates from
Grades 6-11. The estimate for the Grade 7 time score is now 1.38, instead of the original
value which was set at 1 and based on equidistance in time from Grade 6 to Grade 11. This
2
As recommended by Fina (2014), a piecewise LGM was also examined, but did not improve model
fit over the linear LGM.
61
sample accelerates between Grades 6 and 7 more so than would be expected based on the
achievement growth from Grades 7-11. The model estimated correlation coefficient
between the IW and SW was not significant. The disturbance terms decreased, as
expected (Table 4.3). The average standardized disturbance is .31, indicating that the
and 11.
The next step was to examine model fit. Overall, this model was found to be an
improvement over the linear model and to have adequate fit (see Table 4.4). The BIC
index improved by over 60 points from a 46733.82 for the linear LGM to a 46667.56 for
the Free LGM. The RMSEA and SRMR were higher than the recommended cutoffs, but
the CFI was at 0.983. In addition, a comparison of the observed and estimated means
shows that the model reproduced the observed means well (Table 4.5). Covariates will
now be added to this model. The results will be discussed next, along with the results for
Given that appropriate unconditional models were found for Science, Math, and
throughout this study, gender and STEM status (STEM graduate versus non-STEM
major) are both of interest as predictors in the models. Particularly of interest was
whether there differences in the growth of females and males and in the growth of STEM
graduates and non-STEM majors. Also of interest was whether there is an interaction
effect seen for gender and STEM status. Results for these conditional models will be
62
outlined below by subject. The model estimates, fit statistics, and estimated means for
translates to approximately -6 points on the NSS scale. STEM status was also found to be
scale. In addition, the interaction between gender and STEM was not significant for
either IW or SW. This indicates that females started out scoring lower than males in
Science in Grade 6, but did not grow at a significantly different rate than males. Also,
STEM graduates started out scoring higher than non-STEM majors in Science in Grade 6,
but did not grow at a significantly different rate than non-STEM majors through Grade
11. See Figures 4.4-4.6 for graphs of Science means for Grades 6-11 by gender, STEM
status, and both gender and STEM status. A summary of the covariate loadings for all
and a significant predictor of SW (-0.202, p<.01). STEM status was also found to be
significant for both IW (0.167, p<.01) and SW (0.193, p=.01). However, the interaction
between gender and STEM was not significant for either IW or SW. These findings
indicate that females started out at a lower achievement level than males in Math in
Grade 6 and also grew at a slower rate than males from Grades 6-11. Alternately, STEM
graduates started out at a higher achievement level in Grade 6 than non-STEM majors
and grew at a slightly faster rate than non-STEM majors. See Figures 4.7-4.9 for graphs
63
of Math means for Grades 6-11 by gender, STEM status, and both gender and STEM
status.
For Reading, gender was not a significant predictor of IW or SW. STEM was a
significant predictor of IW (0.080, p<.05), but not of SW. The interaction between
gender and STEM was not a significant predictor for either IW or SW. These findings
indicate that, in general, males and females have similar starting achievement levels in
Grade 6 and grow at similar rates from Grade 6-11. While STEM graduates have a higher
majors.
Summary
models that described the growth of students in the subjects of interest. It was found that
there tended to be a lot of variation in individual observed patterns of group and in the
overall starting achievement level; however, the slopes of the overall growth from Grade
6-11 for Reading and Science tended to not have a lot of variability. The second
component, and a key area of interest for this study, was to then examine the effects of
gender and STEM status on these models. For the second component, overall, the results
were as expected for STEM graduates. STEM graduates tended to start at higher
achievement levels for all subjects. While STEM graduates were not found to grow at
faster rates in Science and Reading, they did for Math. Females were found to start at
slightly lower achievement levels in both Science and Math, but had similar starting
achievement levels for Reading in Grade 6. There was not a difference in the rate of
growth for females and males. Nor was there a difference in starting achievement level
64
or rate of growth when considering both gender and STEM status together, i.e. comparing
female STEM students, male STEM students, female non-STEM students, and male non-
STEM students. These findings will be discussed in context as part of Objective 3 when
Benchmarks. As outlined in Chapter III, the benchmarks will link Grade 11 performance
on the IAs in Science and Math to performance in Science and Math courses. The STEM
Readiness Benchmarks represent the level of performance on the IAs that is associated
Therefore, the STEM Readiness benchmarks represent the level of achievement needed
in Grade 11 to be ready for college STEM courses. To begin the process, typical first-
Course Selection
introductory courses for STEM majors. See Appendix A for a list of these courses by
STEM major. Next, the science and math courses students actually enrolled in their first
year were also inspected. As a result, four additional math courses were considered:
Properties of Spaces and Functions II, and Introduction to Linear Algebra. Of these
courses, only Engineering Mathematics II: Multivariable Calculus was added to the
course list. The other three courses had small n-counts and represented just 30 students
for all three courses combined. Therefore, they were not considered typical first-year
65
STEM courses, but rather advanced math courses taken by a select group of students. For
the Math benchmark datasets, the final courses included a variety of Calculus classes. For
the Science benchmark datasets, the final courses included Biology, Chemistry, Physics
and Engineering classes. See Appendix B for a list of the identified first-year STEM
courses. Descriptive statistics for the STEM Readiness Benchmark datasets were then
examined.
Descriptive Statistics
required to have a Grade 11 NSS on the IAs in the subject of interest (i.e., Science or
Math) and to have enrolled in one of the first-year STEM courses at the University of
Iowa in the 2008-2009 school year. In total, there were 931 students in the Science
dataset and 627 students in the Math dataset. While the LGM datasets represent all
students who had the required NSSs and enrolled at the University of Iowa for the 2008-
2009 schoolyear, the benchmark datasets represent only students who had the required
NSS and enrolled in first-year STEM courses. See Table 4.7 for the n-counts by courses
In the Science dataset, 38% were female and 62% were male. Approximately 38%
were STEM graduates and 62% were non-STEM majors. In addition, 27% of the females
were STEM graduates and 45% of the males were STEM graduates. This translates to 95
females and 258 males who were STEM graduates in the Science datasets. In the Math
dataset, 35% were female and 65% were male, and 38% were STEM graduates and 62%
66
STEM graduates and 45% of the males were STEM graduates. This translates to 55
females and 182 males who were STEM graduates in the Math dataset.
NSSs on the IAs and GPAs in high school and college (see Tables 4.8 and 4.9). This is
true when compared to students in the nation (as based on average NPRs) and students
in the LGM datasets. For example, the average NPRs for Grade 11 Science and Math
were both 93 in the benchmark datasets, versus 87 and 86, respectively, in the LGM
datasets. See Table 4.8 for mean NSSs and NPRs in Science and Math on the IAs for
Grades 6-11 in the respective benchmark datasets. Students had an average high school
GPA above a 3.7 and an average college GPA above a 3.0. See Table 4.9 for GPAs in
high school (HSGPA), first year of college (FYGPA), and final college (CGPA) for the
gender, STEM status, and growth patterns will be reported under the Objective 3 section
by subject.
Science
Table 4.10 for the distribution of grades for all Science courses combined. The Science
STEM Readiness Benchmark represents the level of performance on the IAs that is
courses. This is the same level of success utilized in setting ACT’s STEM Readiness
Benchmarks (Mattern et al., 2015), as well as the general CRBs on the IAs (Fina, 2014).
67
Logistic regression estimates the probability of obtaining a particular grade in
science courses based on students’ Science scores on the IAs. For the benchmark
analyses, students’ science grades were coded such that a grade of B or above is 1 and a
grade below a B is 0 (i.e., Stem Ready (SR) = 1 or SR = 0). The regression model was
specified as
The probability of the obtaining a B or above given a particular NSS was expressed as
The results from the logistic regression is summarized in Table 4.11. The results
include a logit parameter and a threshold parameter for each subject. The estimate for the
logit parameter represents a logistic regression coefficient that gives the change in the log
odds of the outcome (SR or STEM Ready) for every one unit increase in the predictor
variable (Grade 11 NSS in Science). Therefore, the change in the log odds of being
STEM Ready is 0.026 for every 1 point increase on the NSS scale in Science. This logit
value is small because there is a large range of possible scores on the IAs NSS scale. The
estimate of the threshold parameter was 8.876. The logit and threshold can be used in
equation 4.2 to determine the cut point in Science. The resulting STEM Readiness
Benchmark was 341 in Science. This score corresponds to a .50 probability of receiving a
B or above in the first-year Science STEM courses. This score also corresponds to an
NPR of 93, meaning that a person receiving this score scored at or above 93% of other
test takers in the nation. An NSS of 341 also corresponds to a raw score of 39/48 items on
the Science test in Grade 11. Therefore, while this is a high NPR, there is not a ceiling
effect for the test as there are still 9 more raw score points above the benchmark.
68
Moreover, this high NPR is not surprising given that the average NPR of students in this
group was also 93 (Table 4.8). The STEM Readiness Benchmark in Science was slightly
higher than the general CRB in Science (341 versus 336; see Table 4.12). There was a 5
point difference on the NSS scale, which corresponds to a 2 raw score points. Whether or
not this was a meaningful difference in the ability of the benchmarks to predict success in
STEM courses will be examined in the Objective 3 section. It should be noted that the
similarity in these values cannot be explained by an overlap in the courses used to set
both benchmarks. The only common course between both benchmarks was Principles of
Benchmark dataset and 15% of the general Science CRB dataset (Fina, 2014). See Table
4.12 for a comparison of the Science STEM Readiness Benchmark to the general CRBs
and their corresponding values on the NPR and raw score scale.
To better understand the Science STEM Readiness Benchmark, the courses used
courses were identified as those most likely to be taken by STEM majors in the first year
of college. Principles of Chemistry I represents about 60% of the dataset and the other
three courses comprise the remaining 40% in almost equal proportions (Table 4.7). See
Table 4.13 for the grade distributions by course. There is variability in the grade
is also variability by course in the percentage of students who ultimately graduated with
a STEM degree. Just under 30% of the students in the Principles of Biology I and
69
Principles of Chemistry I went on to graduate with a STEM degree, while over 60% of
a STEM degree.
Math
combination of math courses identified as those typically taken by STEM majors in the
first year of college, namely specific calculus courses. The logistic methods are the same
The results from the logistic regressions are summarized in Table 4.11. The
estimate for the logit parameter was 0.019. Therefore, the change in the log odds of
being STEM Ready is 0.019 for every 1 point increase on the NSS scale in Math. The
estimate for the threshold parameter was 6.288. As with Science, the logit and threshold
can be used in equation 4.2 to determine the cut point in Math. The resulting STEM
Readiness Benchmark was 331 in Math (Table 4.12). This score corresponds to a .50
probability of receiving a B or above in the first-year Math STEM courses. This score
also corresponds to an NPR of 94, meaning that a person receiving this score scored at
or above 94% of other test takers. An NSS of 331 is equivalent to a raw score of 32/40
items on the Math test in Grade 11. Therefore, while this is a high NPR, there is not a
ceiling effect for the test as there are still 8 more raw score points above the benchmark.
This high NPR is not surprising given the fact that the average NPR of the group was 93
(Table 4.8). It is similar to the results in Science where the average NPR for the group
was also the NPR of the Science Readiness Benchmark. The Math STEM Readiness
70
Benchmark was higher than the general CRB (Fina, 2014) by 12 points on the NSS scale,
which corresponds 4 raw score points. See Table 4.12 for a comparison of the two.
To better understand the Math STEM Readiness Benchmark, the courses used to
establish the benchmark will be examined further. As mentioned previously, five courses
were identified as those taken by STEM majors in the first year of college. The two
largest courses were selected for further examination: Calculus for the Biological
Sciences (represents 30% of the dataset) and Engineering Mathematics I: Single Variable
Calculus (represents 23% of the dataset; see Table 4.7). See Table 4.14 for the grade
distributions by course. Similar to the Science courses, there is variability in the grade
distributions by Math course. For all Math courses combined, approximately 50% of the
students received a grade of B or above (Table 4.10). Alternately, for the two selected
courses, these values were approximately 45% for the biological math class and 40% for
the engineering math class (Table 4.14). These findings, of course, indicate that some of
the remaining Math courses had distributions where over 50% of the students received a
with a STEM major differed widely based on Math course, as it did based on Science
course. Under 15% of the students in the Calculus for Biological Science went on to
graduate with a STEM major, while almost 50% of the students in the Engineering
Summary
As hypothesized, the STEM Readiness Benchmarks were higher than the general
CRBs set by Fina (2014). While the Science STEM Readiness Benchmark was only slightly
higher than the general CRB, the Math STEM Readiness Benchmark reflected a
71
larger increase on the raw score scale as compared to the general CRB in Math. These
findings reflect the fact that students need to be better prepared academically if planning
to be a STEM major. In addition, it was found that there were differences in the grading
practices across the first-year STEM courses for both Science and Math. Objective 3 will
now compare the error rates of the STEM Readiness Benchmarks in Science and Math to
the general CRBs set by Fina (2014). Additional validity evidence for the new
Validity evidence for the STEM Readiness Benchmarks was examined through
multiple methods. First, false-positive error rates, false-negative error rates, and the
percentage of correct classifications (PCC) were calculated. These values were compared
to those found using the general CRBs established by Fina (2014). These values were
also calculated for selected individual Science and Math courses to determine if the
goodness of fit tests were run to determine if there were differences in the error rates
when results were examined for different groups, such as by gender, STEM status, and
growth from Grades 6 to 11. The implications of these findings are discussed in Chapter
V.
Error Rates
the benchmarks to serve as readiness indicators in STEM courses. The results for the
72
Science. For Science the PCC was 61.12. If the general CRBs (Fina, 2014) had
been used to determine readiness in Science STEM courses, the PCC would have been
57.36. Therefore, use of the Science STEM Readiness Benchmark results in greater
accuracy in predicting whether students are ready for Science STEM courses as
compared to using the general CRBs. It is important to note that these differences were
found even though the two types of benchmarks differed by only 2 raw score points.
Compared to the general Science CRB, the Science STEM Readiness Benchmark had a
lower false-positive rate, but a higher rate of false negatives, as expected. Therefore, the
Science STEM Readiness Benchmark is less likely to tell students that they are ready
for STEM coursework when they are actually not (false-positive), but more likely to tell
students that they are not ready for STEM coursework when they actually are (false-
negative).
and false-negatives. This method was not used in this study because this study aimed to
replicate the procedures used by Fina (2014) when establishing the general CRBs on the
IAs and by Mattern et al. (2015) when establishing the STEM Readiness Benchmarks on
the ACT. There are advantages and disadvantages to balancing the error rates of Science
The error rates and PCCs were also calculated by individual Science courses (see
Table 4.16). Overall, these values were similar to the values for the Science STEM
Readiness Benchmark (calculated with all courses) for Principles of Biology I and
Principles of Chemistry I. For Introductory Physics I, the error rates were roughly the
same, except that the rates of false-negatives were lower. For Engineering Problem
73
Solving I, the PCC was higher, the false-negatives were higher, and the false-positives
were lower. The differences across courses can partially be explained by the grade
distributions. For example, in the engineering class over 80% of students received a
grade of B or above. This discrepancy in grading practice from the other courses would
help explain why the false-negatives were higher and the false-positives were lower for
this course. Further conclusions about differences between courses and differences in the
error rates of the STEM Readiness Benchmark across courses cannot be drawn from the
Math. For the Math STEM Readiness Benchmark, the PCC was 54.54. This was
only slightly higher than if the general CRBs had been used (53.11). As expected, the
Math STEM Readiness Benchmark resulted in a lower rate of false-positives and a higher
rate of false-negatives compared to the general CRB. The greatest difference when
comparing the results in Math for use of the Math STEM Readiness Benchmark versus
the general Math CRB was the false-positive rates. If the general Math CRB was used to
predict whether students were ready for Math STEM courses, the false-positive rate
would be 40.19, meaning that approximately 40% of students would be told that they
were ready for Math STEM courses, when they were not. Alternately, the false-positive
rate for the Math STEM Readiness Benchmark was lower at 30.46.
The error rates and PCCs were also calculated by individual Math courses (see
Table 4.17). For Calculus for the Biological Sciences, the false-negatives and PCC were
roughly similar to the values for the Math STEM Readiness Benchmark (calculated with
all courses), but the false-positives were lower. For Engineering Mathematics I: Single
Variable Calculus, the PCC was similar, but the false-negatives were lower and the false-
74
positives were higher. The differences across error rates for courses in Math are not as
easily related to the grade distributions as they were for the Science courses. Further
conclusions about differences between courses and differences in the error rates of the
Math STEM Readiness Benchmark across courses cannot be drawn from the existing
results and are outside the scope of this study. That said, the influence of grading
practices and sources of variance, other than test scores, that affect whether college
students are successful in Math and Science courses will be highlighted in Chapter V.
Now that the capability of the STEM Readiness Benchmarks to serve as readiness
indicators in STEM courses has been considered, the benchmarks will also be examined
to see if there are differences based on gender and achievement growth patterns. In
addition, validity evidence for the STEM Readiness Benchmarks for predicting another
measure of STEM success, namely graduating with a STEM degree will be examined.
A chi-square goodness of fit test can be used to determine how well observed
values of the data match the expected values of the data using two-by-two tables.
Χ2 = ∑ (observed-expected)2. [4.3]
expected
The value of the test statistics is compared to the value of the null hypothesis, Χ2 with k-
1 degrees of freedom where k is 4. Then the observed values are examined to see which
values are not as expected. In this application, the expected values are the percentages
obtained from the entire group. Researchers often visually inspect the data and make a
judgement call as to which variables contributed to the test statistic; however Sharpe
75
(2015) offers several follow-up methods to empirically evaluate the data, including
calculating residuals.
Follow-up test methods. For this study, standardized residuals are calculated for
each cell. A standardized residual is calculated by taking the difference in the observed
and expected values and dividing this difference by the square root of the expected
value. Sharpe (2015) does not offer a cutoff for evaluating the magnitude of standardized
residuals, but cites several researchers that recommend that the cells of the larger
the magnitude of adjusted standardized residuals when the number of cells is small;
however calculating adjusted standardized residuals is not an option for this study as
those can only be calculated for tables larger than two-by-two. Considering these
recommendations, cells in which the absolute value of the standardized residual is above
For this study, the expected values are the error rates and PCC for the STEM
Readiness Benchmarks in Science and Math. The observed values are the error rates and
PCC when the benchmarks are applied considering gender and growth. That is, the
STEM Readiness Benchmarks will be examined to see if they function differently for
females and males, for students who demonstrate a more rapid rate of growth from
Grade 6 to Grade 11 versus students who do not show as much growth, and for STEM
graduates. The expected values for the error rates and PCC can be found in Table 4.15;
however, this table does not separate the PCC. The PCC consists of students who were
1) correctly classified as being ready for STEM coursework, i.e, they met the STEM
Readiness Benchmark and they received a B or above in the course and 2) correctly
76
classified as not being ready for STEM coursework, i.e., they did not meet the STEM
Readiness Benchmark and they received below a B in the coursework. Students who
were correctly classified as being ready for STEM coursework will now be categorized
as PCC Yes and students who were correctly classified as not being ready for STEM
coursework will now be categorized as PCC No. For Science, the PCC Yes was
approximately 38 and the PCC No was approximately 23 (total PCC = 61.12 from Table
4.9). For Math, the PCC Yes was approximately 34 and the PCC No was approximately
Defining growth groups. To examine how growth in Science and Math on the
IAs is related to the STEM Readiness Benchmarks, students in the STEM Readiness
Benchmark datasets were categorized into one of three growth groups: low, medium, and
high. Growth groups were defined by examining the distribution of the change in NSS
from Grade 6 to Grade 11 (Grade 11 NSS – Grade 6 NSS = change in NSS). Changes in
NSS in Science ranged from -11 to 134. Changes in NSS in Math ranged from 23 to 132.
Some students in the benchmark datasets were missing Grade 6 NSSs and were omitted
from the analyses (388 students in Science, 233 students in Math). Because there were no
natural breaks detected in the distribution of change scores and because the group showed
more growth, on average, than would be expected based on NSSs in the nation, it was
decided to create groups so that roughly one third of students were in each group. For
Science, low growth was defined as a change in NSS below 67 points, medium growth
from 67-86 points, and high growth was above 86 points. For Math, low growth was
defined as a change in NSS below 67 points, medium growth from 67-82 points, and high
growth was above 82 points. It should be noted that low growth is only low relative
77
to this group of students. Most students in the low growth groups are still above the
average growth for students in the nation. Expected average change in NSS in the nation
from Grade 6 to Grade 11 is approximately 48 points. Only about 10% of students in the
Science and Math benchmark datasets fall below the national expected growth.
Results of the chi-square goodness of fit tests for gender, stem status, and growth
groups are summarized in Table 4.18. For Science, the following groups showed
differences in the distribution of error rates and PCCs as compared to those for the total
group: STEM graduates, non-STEM majors, and students in both the low growth group
and high growth group. For Math, the following groups showed differences: females,
males, STEM graduates, non-STEM majors, and students in the low growth group.
STEM. Follow-up tests revealed that both STEM Readiness Benchmarks were
correctly classified as being ready for STEM coursework and were also less likely to be
correctly classified as not being not ready. Conversely, non-STEM majors were less
likely to be correctly classified as being ready for STEM coursework and more likely to
Gender. Follow-up tests did not find differences in the error rates for the Science
STEM Readiness Benchmark by gender, but did find differences in the error rates for the
Math STEM Readiness Benchmark by gender. Specifically, the Math STEM Readiness
Benchmark was more likely to give females false-negatives and less likely to give them
false-positives than males. Overall, these findings indicate that the Science STEM
Readiness Benchmark functions similarly for females and males; while the Math STEM
78
more likely to be told they are not ready for STEM classes in Math when they
actually are and less likely to be told that they are ready when they actually are not.
These findings are important to consider when counseling high school students and
Growth group. Results of the follow-up tests on the growth groups also differed
for Science and Math. The Science STEM Readiness Benchmark was more likely to
correctly classify students in the low growth group as not ready for STEM coursework,
less likely to correctly classify them as ready, and more likely to give them false-
negatives. For the high growth group, the results for the PCCs were reversed and the
Science STEM Readiness Benchmark was less likely to give false-negatives and more
likely to give false-positives. The Math STEM Readiness Benchmark only functioned
differently for the low growth group. For the low growth group, the Math STEM
Readiness Benchmark was more likely to correctly classify students as not ready and less
These findings indicate that students who demonstrate an especially large amount
STEM courses, but would also be more likely to get a false-positive from the Science
STEM Readiness Benchmark. In addition, students who fall into the low growth group
would be less likely to be successful in Science and Math STEM courses, but also more
status. To better understand the composition of the different groups, mean NSSs in
Grade 11 in Science and Math on the IAs by gender, STEM, growth group status, and
79
benchmark status can be found in Tables 4.19 and 4.20, for Science and Math
respectively. For the growth groups, means for Grade 6 are also included, as the groups
were defined based on Grade 6 and Grade 11 scores. For all others, only a Grade 11
score was required to be a part of these datasets, therefore, many students are missing
the other grades. Mean ACT scores by benchmark status can be found in Table 4.21 for
Overall, males scored higher than females, STEM graduates scored higher than
non-STEM majors, and those who met the STEM Readiness Benchmarks scored higher
(by definition). The means by growth group status show some interesting trends. The low
growth group had the highest means in Grade 6, but the lowest means in Grade 11.
Alternately, the high growth group had the lowest means in Grade 6 and the highest
means in Grade 11. While these groups were defined by the amount of growth shown
from Grade 6 to Grade 11, these results, especially the magnitude of the differences
between the groups’ scores, were not expected. That is, it was not expected that the low
group would necessarily have the highest Grade 6 NSS and the lowest Grade 11 NSS.
Nor were the large differences expected between the Grade 11 scores for low group and
the high group. These findings may help to explain the results from the chi-square
goodness of fit tests. For example, as the high growth groups tended to have higher
Science and Math scores in Grade 11, they would have tended to meet the benchmark
more often. Thus, it follows that they would also be more likely to receive false-positives
and less likely to receive false negatives. The reverse would be true for the low growth
groups.
80
Summary
The STEM Readiness Benchmarks for both Science and Math were found to be
higher than the general CRBs set by Fina (2014). They were also found to be more
accurate in predicting success in first-year STEM courses than if the general CRBs were
predicting STEM graduates. Differences were found in the grade distributions of the
individual courses that defined the benchmarks. Furthermore, there were differences in
the error rates and PCCs when calculated by individual courses. These findings highlight
the effects of grading practices on success in STEM courses. They also highlight how
course selection for setting the benchmarks may influence the benchmarks. Implications
Next, the STEM Readiness Benchmarks were examined to see if there were
differences by gender and growth group. Differences were found by gender for the Math
STEM Readiness Benchmark, namely that this benchmark result in more false-negatives
for females and less false-positives. Differences were also found by growth group status.
These differences may be explained, at least in part, by the large mean differences
between the groups. However, it also appears that the Science STEM Readiness
Benchmark results in more false-negatives for the low growth group and less false-
Summary
The chapter summarized the results for the three objectives. Objective 1
identified models that described the growth of students in Science, Math, and Reading,
and then described differences by gender and STEM status. Overall, males and STEM
81
graduates tended to start at higher levels of achievement in Grade 6, but they did not tend
to grow at different rates than females and non-STEM majors. Objective 1 aimed to
better understand STEM graduates achievement growth throughout middle school and
high school.
Objective 2 established STEM Readiness Benchmarks on the IAs for Math and
Science. The STEM Readiness Benchmarks were higher than the general CRBs,
reflecting the fact that students need to be better prepared academically if planning to
major in STEM. Differences in grading practices for the individual STEM courses were
found.
The STEM Readiness Benchmarks were found to provide a more accurate picture of a
student’s readiness for a STEM major than general CRBs. In addition, they were found
described for gender and growth patterns. Chapter V will summarize trends and discuss
the implications of these findings for students and educators. In addition, limitations of
this study will be discussed and suggestions for future research will be offered.
82
Table 4.1
Means (SDs) of GPAs on the LGM Datasets
All Students
Group N Mean (SD) Min Max
All HSGPA 1346 3.59 (0.36) 2.42 4.46
FYGPA 1264 3.02 (0.66) 0.06 4.13
CGPA 1137 3.07 (0.55) 0.46 4.12
By Gender
Group N Mean (SD) Min Max
Female HSGPA 719 3.63 (0.33) 2.49 4.38
FYGPA 663 3.06 (0.66) 0.39 4.11
CGPA 590 3.12 (0.52) 1.39 4.12
Male HSGPA 627 3.54 (0.37) 2.42 4.46
FYGPA 601 2.99 (0.67) 0.06 4.13
CGPA 547 3.01 (0.58) 0.46 4.07
By STEM Status
Group N Mean (SD) Min Max
STEM HSGPA 144 3.83 (0.27) 2.89 4.46
FYGPA 144 3.42 (0.45) 2.25 4.13
CGPA 144 3.34 (0.44) 2.07 4.08
Non-STEM HSGPA 1202 3.56 (0.35) 2.42 4.42
FYGPA 1120 2.97 (0.67) 0.06 4.11
CGPA 993 3.03 (0.55) 0.46 4.12
By STEM Status and Gender
Group N Mean (SD) Min Max
STEM
Female HSGPA 42 3.88 (0.17) 3.39 4.19
FYGPA 42 3.46 (0.47) 2.33 4.08
CGPA 42 3.33 (0.46) 2.07 4.08
STEM
Male HSGPA 102 3.81 (0.3) 2.89 4.46
FYGPA 102 3.40 (0.44) 2.25 4.13
CGPA 102 3.34 (0.44) 2.18 4.07
Non-STEM
Female HSGPA 677 3.62 (0.34) 2.49 4.38
FYGPA 621 3.03 (0.66) 0.39 4.11
CGPA 548 3.10 (0.52) 1.39 4.12
Non-STEM
Male HSGPA 525 3.48 (0.36) 2.42 4.42
FYGPA 499 2.90 (0.68) 0.06 4.07
CGPA 445 2.93 (0.58) 0.46 4.06
83
Table 4.2
Means (SDs) of Scores on the Iowas for LGM Datasets
Science
Grade N NSS NPR
6 1128 263.00 (27.09) 81.21 (16.56)
7 1283 278.39 (28.73) 81.43 (16.87)
8 1327 288.99 (26.90) 80.55 (15.52)
11 1327 332.35 (28.70) 87.21 (13.67)
Math
Grade N NSS NPR
6 1221 252.67 (22.54) 80.04 (17.78)
7 1302 271.51 (22.06) 82.77 (15.57)
8 1344 286.55 (22.86) 83.48 (15.48)
11 1344 321.20 (23.88) 85.96 (14.21)
Reading
Grade N NSS NPR
6 1222 254.05 (25.61) 78.33 (18.32)
7 1303 272.35 (28.68) 80.30 (18.29)
8 1346 282.20 (27.11) 78.75 (17.29)
11 1346 321.70 (31.55) 82.18 (15.68)
84
Table 4.3
Model Estimates for the Latent Growth Models in Science, Math, and Reading
Linear 263.02 13.77 493.19 2.08b -7.76b -0.24a 279.82 297.90 272.54 358.57 0.36 0.38 0.37 0.43
b a
Cond. 263.03 13.77 474.76 2.05 -9.57 -0.31a 279.51 296.45 274.46 353.87 0.36 0.38 0.37 0.43
Math
Linear 256.77 13.40 391.78 0.88b -2.75b -0.15b 155.04 107.52 118.10 208.77 0.28 0.22 0.24 0.35
a
Free 253.33 16.87 399.74 4.39 -7.10 -0.17 120.10 111.69 103.60 177.16 0.23 0.22 0.21 0.30
Cond. 253.34 16.85 366.04 4.20 -11.55 -0.30 119.65 112.13 104.65 171.95 0.23 0.22 0.21 0.29
Reading
Linear 255.98 13.31 539.72 0.79b 0.00 0.00 199.40 207.78 191.94 454.39 0.27 0.28 0.26 0.45
b b b
Free 254.28 13.61 530.49 1.15 3.16 0.13 191.35 194.28 194.56 449.20 0.27 0.26 0.26 0.43
Cond. 254.28 13.61 524.90 0.86b 2.03b 0.10b 192.91 194.05 195.50 440.92 0.27 0.26 0.26 0.43
Note: All values are significant at the .01 level unless indicated otherwise.
a
Significant at the .05 level. b Not significant at the .05 level.
Table 4.4
Fit Statistics for the Latent Growth Models in Science, Math, and Reading
Free 90% CI
Model Par BIC CFI RSMEA Lower Upper SRMR
Science
Linear 9 46030.54 0.985 0.069 0.049 0.091 0.098
Conditional 15 45995.02 0.985 0.047 0.033 0.062 0.064
Math
Linear 9 44092.15 0.909 0.224 0.204 0.244 0.096
Free 10 43806.30 0.988 0.091 0.070 0.115 0.140
Conditional 16 43649.97 0.988 0.059 0.045 0.074 0.084
Reading
Linear 8 46733.82 0.958 0.127 0.109 0.146 0.163
Free 10 46667.56 0.983 0.098 0.076 0.122 0.197
Conditional 16 46664.12 0.971 0.083 0.069 0.098 0.126
85
Table 4.5
Observed and Estimated Means for the Latent Growth Models in Science, Math, and Reading
Table 4.6
Covariate Loadings for the Conditional Latent Growth Models in Science, Math, and Reading
Intercept Slope
Model (IW) (SW)
Science
Gender -0.133 NS
STEM 0.120 NS
Interaction (Gender*STEM) NS NS
Math
Gender -0.205 -0.202
STEM 0.167 0.193
Interaction (Gender*STEM) NS NS
Reading
Gender NS NS
STEM 0.080a NS
Interaction (Gender*STEM) NS NS
Note: All values are significant at the .01 level unless indicated otherwise.
a
Significant at the .05 level. NS is not significant at the .05 level.
86
Table 4.7
N-Counts by Course for the STEM Benchmarks
Table 4.8
Means (SDs) of Scores on the Iowas for STEM Benchmark Datasets
Science
Grade N NSS NPR
6 543 271.44 (24.41) 86.20 (12.86)
7 623 289.21 (24.74) 87.44 (12.44)
8 654 298.53 (24.48) 85.84 (12.51)
11 931 346.62 (22.46) 93.06 (8.31)
Math
Grade N NSS NPR
6 411 260.08 (20.71) 85.64 (14.95)
7 443 279.75 (18.53) 88.61 (11.41)
8 465 296.43 (18.62) 89.81 (10.48)
11 657 334.24 (18.71) 92.89 (8.99)
87
Table 4.9
Means (SDs) of GPAs on the Benchmark Datasets
Science
Group N Mean (SD) Min Max
All HSGPA 929 3.74 (0.32) 2.60 4.61
FYGPA 922 3.18 (0.64) 0.00 4.18
CGPA 850 3.12 (0.57) 1.43 4.13
Math
Group N Mean (SD) Min Max
All HSGPA 657 3.71 (0.32) 2.60 4.46
FYGPA 648 3.13 (0.66) 0.00 4.18
CGPA 596 3.08 (0.57) 1.43 4.09
Table 4.10
Cumulative Percent Frequency Distribution of Grades in each Subject
88
Table 4.11
Results from the Logistic Regression Models
Table 4.12
Values for the STEM Readiness Benchmarks versus General CRBs
Table 4.13
Cumulative Percent Frequency Distribution of Grades in Science Courses
Course Grade Prin of Bio I Prin of Chem I Intro Physics I Eng Prob Solv I
A+ 5.51 0.73 0.81 3.79
A 25.98 7.85 9.68 32.58
A- 25.98 16.61 18.55 48.48
B+ 33.07 26.64 29.84 66.67
B 60.63 40.15 48.39 81.82
B- 61.42 55.29 61.29 88.64
C+ 67.72 69.71 69.35 90.15
C 89.76 81.39 77.42 93.18
C- 89.76 87.23 90.32 93.94
D+ 89.76 90.15 92.74 94.70
D 93.70 92.15 95.97 94.70
D- 93.70 93.25 95.97 95.45
F 96.06 95.99 96.77 96.21
Incomplete 96.06 96.17 97.58 96.97
Withdrawn 100.00 100.00 100.00 100.00
Note: A grade of B is the cutoff for being considered STEM Ready.
89
Table 4.14
Cumulative Percent Frequency Distribution of Grades in Selected Math Courses
90
Table 4.15
Error Rates for the STEM Readiness Benchmarks versus General CRBs
Table 4.16
Error Rates for the Science STEM Readiness Benchmark by Science Course
Table 4.17
Error Rates for the Math STEM Readiness Benchmark by Selected Math Course
91
Table 4.18
Chi-Square Goodness of Fit Test Results
Χ2
Test False- False- PCC PCC
Group df Stat Negative Positive No Yes
Science 12 27 23 38
Female 3 7.00 11.48 24.37 28.85 35.29
Male 3 3.20 12.54 28.22 19.86 39.37
STEM 3 59.44 11.90 20.11 11.33* 56.66*
Non-STEM 3 38.56 12.28 30.80 30.62* 26.30*
Low Growth 3 25.80 17.78* 22.22 35.00* 25.00*
Med Growth 3 2.19 10.00 27.78 20.00 42.22
High Growth 3 23.78 4.92* 35.52* 13.11* 46.45*
Math 15 30 21 34
Female 3 18.26 21.00* 18.72* 26.94 33.33
Male 3 11.26 11.76 36.76 17.65 33.82
STEM 3 34.11 12.66 27.43 9.70* 50.21*
Non-STEM 3 21.96 16.41 32.31 27.69* 23.59*
Low Growth 3 22.31 19.23 24.62 35.38* 20.77*
Med Growth 3 9.38 6.92 33.08 17.69 42.31
High Growth 3 6.85 15.67 32.84 11.94 39.55
*Indicates an absolute value of standardized residual approximately 1.5 or larger.
92
Table 4.19
Means (SDs) of Scores on the Iowas for the Science STEM Readiness Benchmark Dataset
By Gender
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Female 357 -- 342.50 (23.25)
Male 574 -- 349.18 (21.59)
By STEM Status
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
STEM 353 -- 353.73 (19.22)
Non-STEM 578 -- 342.23 (23.09)
By Growth Group
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Low 180 285.84 (22.58) 336.74 (24.03)
Medium 180 272.37 (21.70) 349.46 (21.11)
High 183 256.35 (19.36) 356.07 (15.56)
By Science STEM Readiness Benchmark Status
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Met Benchmark 601 -- 360.44 (10.07)
Did Not Meet Benchmark 330 -- 321.44 (16.01)
93
Table 4.20
Means (SDs) of Scores on the Iowas for the Math STEM Readiness Benchmark Dataset
By Gender
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Female 219 -- 327.81 (19.95)
Male 408 -- 337.69 (17.06)
By STEM Status
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
STEM 237 -- 341.60 (15.83)
Non-STEM 390 -- 329.77 (18.93)
By Growth Group
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Low 130 270.11 (18.19) 324.88 (20.80)
Medium 130 264.11 (16.20) 337.96 (15.64)
High 134 246.45 (19.76) 340.01 (15.07)
By Math STEM Readiness Benchmark Status
Grade 6 Grade 11
Group N Mean (SD) Mean (SD)
Met Benchmark 402 -- 345.32 (9.75)
Did Not Meet Benchmark 225 -- 314.44 (13.94)
94
Table 4.21
Means (SDs) of ACT Scores for STEM Readiness Benchmark Datasets
95
Random Sample of Patterns of Growth in Science
with Population Mean
375
350
325
300
Population
NSS Scale
275 Mean
250
225
200
175
6 7 8 11
Grade
Figure 4.1 A random sample of 30 students’ patterns of growth in Science with the population
mean.
96
Random Sample of Patterns of Growth in Math
with Population Mean
375
350
325
300
NSS Scale
275 Population
Mean
250
225
200
175
6 7 8 11
Grade
Figure 4.2 A random sample of 30 students’ patterns of growth in Math with the population
mean.
97
Random Sample of Patterns of Growth in Reading
with Population Mean
400
375
350
325
NSS Scale
300
275
Population
250 Mean
225
200
175
6 7 8 11
Grade
Figure 4.3 A random sample of 30 students’ patterns of growth in Reading with the
population mean.
98
Random Sample of Patterns of Growth in Science
with Gender Means
375
350
325
300
NSS Scale
275 Male
250 Female
225
200
175
6 7 8 11
Grade
Figure 4.4 A random sample of 30 students’ patterns of growth in Science with the population
gender means.
99
Random Sample of Patterns of Growth in Science
with STEM Means
375
350
325
300
NSS Scale
275 STEM
Non-
250 STEM
225
200
175
6 7 8 11
Grade
Figure 4.5 A random sample of 30 students’ patterns of growth in Science with the population
STEM means.
100
STEM and Gender Means in Science
375
350
325
NSS
300
275
250
6 7 8 11
Grade
101
Random Sample of Patterns of Growth in Math
with Gender Means
375
350
325
300
NSS Scale
275 M
Male
250
Female
225
200
175
6 7 8 11
Grade
Figure 4.7 A random sample of 30 students’ patterns of growth in Math with the population
gender means.
102
Random Sample of Patterns of Growth in Math
with STEM Means
375
350
325
300
NSS Scale
275
STEM
250 Non-
STEM
225
200
175
6 7 8 11
Grade
Figure 4.8 A random sample of 30 students’ patterns of growth in Science with the population
STEM means.
103
STEM and Gender Means in Math
350
325
300
NSS
275
250
225
6 7 8 11
Grade
104
Random Sample of Patterns of Growth in Reading
with Gender Mean
400
375
350
325
NSS Scale
300
275 Female
250
Male
225
200
175
6 7 8 11
Grade
Figure 4.10 A random sample of 30 students’ patterns of growth in Reading with the
population gender means.
105
Random Sample of Patterns of Growth in Reading
with STEM Mean
400
375
350
325
NSS Scale
300
275
STEM
250 Non-
STEM
225
200
175
6 7 8 11
Grade
Figure 4.11 A random sample of 30 students’ patterns of growth in Reading with the
population STEM means.
106
STEM and Gender Means in Reading
350
325
300
NSS
275
250
225
R6 R7 R8 R11
Grade
107
CHAPTER V: DISCUSSION
Chapter V summarizes and discusses the results presented in Chapter IV. First,
the study objectives are recapped. Next, similarities and differences in achievement
growth on the Iowas (IAs) of successful STEM majors compared to non-STEM majors
are summarized. Gender differences of these two groups are discussed. Then, STEM
differences and the influences of grading practices are highlighted. Further implications
of these finding for students and educators are considered. Finally, limitations of this
Study Objectives
need for STEM graduates; however, there are issues that contribute to a lack of STEM
graduates. These issues include a gender discrepancy in the number of female STEM
graduates and a high rate of attrition of STEM majors. The attrition is due, in part, to
students not being prepared for STEM coursework. These concerns are combined with a
lack of information on the early achievement growth patterns of STEM graduates and a
lack of accurate readiness indicators for first-year STEM courses. This study addressed
these issues through three objectives: Objective 1: Comparing the achievement growth of
Benchmarks in Science and Math on the IAs, and Objective 3: Providing validity
108
Achievement Growth
throughout middle school and high school. To achieve this goal, latent growth models
(LGMs) were used to describe the growth of students on the IAs in Science, Math, and
Reading. Next, the effects of STEM status (i.e., STEM graduate or non-STEM major1)
Results from the LGM analyses demonstrated that STEM graduates start at
higher achievement levels in Grade 6 and maintain higher achievement levels through
Grade 11 in all three subjects studied. The magnitude of the difference in observed
starting achievement levels between STEM graduates and non-STEM majors was largest
in Math and Science, while a smaller difference was found in Reading. This finding was
not surprising given that math and science achievement are positively associated with
selecting a STEM major (Davidson et al., 2014; Wang, 2013) and verbal achievement is
negatively associated with selecting a STEM major (Davidson et al., 2014). Also, when
verbal achievement is higher than math achievement, students are more likely to select a
non-STEM major (Davidson et al., 2014). Therefore, it is probable that many of the high
achieving students in Reading selected a non-STEM major upon entering college and
While STEM graduates tend to start at higher achievement levels compared to non-
STEM majors, they do not grow at different rates from Grade 6 to Grade 11, except
1
STEM graduate and STEM major are used interchangeably in Chapters IV and V to indicate a student
who graduated with a STEM major. Non-STEM majors were classified to include both graduates and
students who may not have graduated, because the purpose of this study was to compare STEM graduates
with those who did not graduate with a STEM degree, whether that was because they graduated with a non-
STEM degree or because they left the university.
109
in Math. In Math, STEM graduates’ achievement levels were found to change at a faster
rate than non-STEM majors’ achievement levels. These findings show that achievement
level differences between STEM graduates and non-STEM majors can be found as early
as Grade 6. Consequently, educators may be able to identify students who would likely
who are interested in STEM careers, but are not on-track for a STEM degree may need to
Gender Differences
slightly lower achievement levels in Grade 6 in Science and Math and slightly higher
levels in Reading. While females start at different achievement levels than males, they
do not change at different rates than males from Grade 6 to Grade 11. In addition, no
interaction effect was found between STEM status and gender. It is important to note the
differences between males and females are not as large as the differences found between
STEM graduates and non-STEM majors. In addition, the achievement level differences
between male STEM graduates and female STEM graduates is minimal. This suggests
that students with high achievement levels self-select as STEM majors, regardless of
gender. It has been shown that prior achievement in science and math does not account
for gender differences in the number of STEM majors (Riegle-Crumb et al., 2012).
Therefore, it is likely that these gender differences in achievement levels did not account
for the gender differences in the number of STEM majors in this sample. Rather, other
factors likely played a role in the observed gender discrepancy in STEM graduates.
Unfortunately, factors that lead to a gender discrepancy in STEM graduates are not well
110
understood and warrant further investigation (e.g., Gayles & Ampaw, 2011; Huang et al.,
benchmarks provide high school students with an accurate indicator of readiness for first-
year STEM coursework in Science and Math. Objective 3 provided validity evidence for
the STEM Readiness Benchmarks in Science and Math. Since these two objectives are
succeed in typical general education courses taken during the first year of college, the
courses taken during the first year of college. It was hypothesized that general CRBs may
not be suitable indicators of readiness for students planning to pursue a STEM major, due
to the fact that STEM majors tend to take more advanced math and science courses than
non-STEM majors (Ost, 2010; Phillip et al., 2005; Rask, 2010, Stinebrickner
& Stinebrickner, 2014). As predicted, the STEM Readiness Benchmarks were higher
than the general CRBs set on the IAs by Fina (2014). Specifically, the Science STEM
Readiness Benchmark was higher than the general Science CRB (341 versus 336 on the
National Standard Score [NSS] scale), and the Math STEM Readiness Benchmark was
higher than the general Math CRB (331 versus 319 on the NSS scale). These results
were consistent with those of Mattern et al. (2015) who found ACT’s STEM Readiness
111
findings reflect that students need to be better prepared academically in science and
Next, validity evidence for the STEM Readiness Benchmarks was provided
through examining the error rates and percentage of correct classifications (PCC) of the
STEM Readiness Benchmarks for predicting success in first-year STEM courses. These
values were first compared to those found using the general CRBs established by Fina
(2014). The STEM Readiness Benchmarks were found to be more accurate in predicting
success in first-year STEM courses than if the general CRBs on the IAs were used, as
measured by the PCC. Both the Science STEM Readiness Benchmark and the Math
STEM Readiness Benchmark had a lower false-positive rate and a higher false-negative
rate compared to the general CRBs. This trend was as expected since the STEM
STEM coursework. Nonetheless, the error rates for Math are substantial and warrant
further discussion.
As mentioned previously, if the general CRB had been used to predict success in
Math STEM courses, the false-positive rate would have been high. While the false-
positive rate was lower when using the Math STEM Readiness Benchmark, it was still
relatively high. However, the false-negative rate for the Math STEM Readiness
positives, benchmarks are often set using a method in which the rates of false-negatives
and false-positives are balanced. While this method does have the advantage of
balancing the error rates, it results in a different value for the benchmark than when using
logistic regression. For instance, if the error rates were balanced for the Math STEM
112
Readiness Benchmark, the benchmark would be raised in order to lower the false-positive
rate. This, of course, would also raise the false-negative rate. Logistic regression was
chosen for this study because it aimed to replicate the procedures used by Fina (2014)
and Mattern et al. (2015) when establishing the general CRBs on the IAs and the STEM
one to make probability statements based on the benchmarks. Instead of altering the
these findings be taken into account when considering the STEM Readiness Benchmarks
negatives have different implications in practice. For example, the cost of a benchmark
with a high rate of false-negatives may be that students who do not meet the benchmark,
decide to not pursue STEM coursework, even though they actually would be successful.
pursue STEM coursework even though they actually may not be successful (as
measured by receiving a grade of B or above in the course). For the purpose of this
study, the rates of both are important to consider in the context of counseling students
may function differently for different groups, the STEM Readiness Benchmarks results
were examined by STEM status, growth group status, and gender. To accomplish this,
error rates were calculated by group and the results were compared through chi-square
113
goodness of fit tests. These findings are summarized below and have important implications
for students and educators that will be discussed in the implications section.
STEM Graduates
Students who met the STEM Readiness Benchmarks were more likely to
graduate with a STEM degree. The overall PCC was significantly higher for STEM
graduates as compared to the PCC for all students. In addition, the PCC Yes was
significantly higher and the PCC No was significantly lower for STEM graduates. These
trends were true for both the Science and Math STEM Readiness Benchmarks.
Growth group status was defined by equally dividing students into one of three
growth groups (low, medium, or high) based on their achievement growth on the IAs
from Grade 6 to Grade 11. On average, the low growth group demonstrated a higher
compared to the other groups. In contrast, the high growth group demonstrated a lower
level of achievement in Grade 6, but a higher level of achievement in Grade 11. Notably,
the low growth group is still well above average in both overall growth from Grade 6 to
Differences in the error rates for the STEM Readiness Benchmarks were found by
growth group status. Overall, the Math STEM Readiness Benchmark functioned fairly
similarly for all growth groups. The only differences were in the low growth where
students were found to be less likely to be ready for Math STEM coursework, as
measured by the Math STEM Readiness Benchmark and their course grades. The Science
STEM Readiness Benchmark functioned differently for both the low growth
114
group and the high growth group. In the low growth group, students were less likely to be
ready for Science STEM coursework as measured by the Science STEM Readiness
Benchmark and their course grades. The Science STEM Readiness Benchmark was also
more likely to give students in the low growth group false-negatives. Alternately,
students who demonstrated an especially large amount of growth from Grade 6 to Grade
11 (high growth group) were more likely to meet the Science STEM Readiness
Benchmark and to be successful in Science STEM courses. They were also more likely to
While these findings are interesting, they require more follow-up before being
generalized to other samples. As mentioned, the groups were defined not by pre-
determined levels of growth, but by dividing the sample into three equal parts.
Therefore, it would be important to validate these findings with other samples to ensure
that the levels of growth used to define the growth groups are meaningful.
Gender Differences
There were not significant differences in the error rates for the Science STEM
Readiness Benchmark by gender; however differences were found for the Math STEM
Readiness Benchmark. The Math STEM Readiness Benchmark resulted in more false-
negatives and less false-positives for females. It is important to note that these
differences found on the Math STEM Readiness Benchmark cannot be accounted for
solely due to gender differences in achievement levels. Females were found to have
lower achievement levels in both Science and Math in Grade 11, but these differences did
not translate to differences in the error rates for the Science benchmark.
115
Grading Practices
used to establish the benchmarks were examined. All of the identified STEM courses are
offered through the College of Liberal Arts and Sciences (CLAS) at the University of
Iowa, except for the three engineering courses which are offered through the College of
Engineering. While the College of Engineering does not have recommended grade
distributions for its courses, the CLAS does (CLAS recommended grade distributions,
2017). These guidelines are the same for all first-year courses and are as follows: 15% A,
34% B, 40% C, 8% D, and 3% F. Even though these guidelines exist, differences were
found in the observed grade distributions of Iowa students across the first-year Science
and Math STEM courses. The percentages of students who received a grade of B or
above ranged from 40% to 80%, depending on the course. Differences were also found in
the error rates and PCC for individual courses. These findings are based on a subsample
of the students who took each course, as data were only available for students who
graduated from high school in Iowa. In the 2008-2009 school year, approximately 54% of
first-year students at the University of Iowa were Iowa residents (University of Iowa,
2008).
The purpose of this study was to provide information related to the levels of
Accurate indicators of STEM readiness are vitally important, as being unprepared for
STEM coursework is associated with STEM attrition and dropping out of college (Chen,
2013; Stinebrickner & Stinebrickner, 2014; Westrick, 2015). If students are being
116
encouraged to pursue STEM majors, it is important they do so armed with as much
encourage students who are not academically prepared to pursue STEM majors, as
switching majors or dropping out of college are costly consequences. The STEM
As discussed in Chapter II, benchmarks are useful tools for quickly assessing
because many factors influence student success in college courses. Other sources of
variance in determining course grades include students’ study skills, time management
skills, motivation, access to campus resources, and relationships with faculty. With this
caveat in mind, this section will discuss additional implications of the findings from this
study and make recommendations for utilizing the STEM Readiness Benchmarks for
Counseling Students
assessing STEM readiness. Students who met the benchmarks were more likely to be
successful in STEM courses in college and were also more likely to graduate with STEM
degrees. For these reasons, the benchmarks could be used to help counsel junior high and
high school students. For example, if a student met the benchmarks, but was not
considering a STEM major, the student could be encouraged to explore STEM options.
This practice would help to address the concern that many students who are talented in
117
student was interested in a STEM major, but did not meet the STEM Readiness
Benchmarks (or was not on-track to meet the benchmarks in Grade 11), remedial work
important to assess and intervene early. Because STEM graduates were found to show
more achievement growth in Math from Grade 6 to Grade 11, it would be particularly
Since differences were found for the STEM Readiness Benchmarks based on the
amount of growth shown from Grade 6 to Grade 11, a guidance counselor could also take
into account the amount of achievement growth a student displayed throughout middle
school and high school. Findings from this study suggest that students who were in the
high growth group from Grade 6 to Grade 11 would be more likely to be successful in
Gender differences. Differences were found in the error rates of the Math STEM
Readiness Benchmark when considering gender. Namely, females were more likely to
receive false-negatives. Approximately 21% of females would be told that they were not
ready for Math STEM coursework when they actually were ready. As mentioned previously,
when counseling students, it is important to consider other factors in addition to test scores.
When counseling female students considering a Math career, it may be especially important
to consider factors such as grades in high school, previous math courses, and motivation. For
example, consider a female high school student who is interested in a Math career and does
118
took advanced math courses in high school and has a solid GPA. This student would be
told she was not ready for Math STEM coursework based on the Math STEM Readiness
Benchmark; however, she may be one of the false-negatives and have a good chance of
being successful in STEM courses. Therefore, the STEM Readiness Benchmarks should
essential to recognize that no test score or cut score interpretation is flawless, but should
study via discussion of the benchmarks’ error rates. In addition, the STEM Readiness
Benchmarks can be presented to students and educators as the score at which 50% of
Curriculum Design
The findings from this study could also be used to impact curriculum design in
science and math. The curriculum for science and math in Grades K-12 are set based on
national standards, specifically the Next Generation Science Standards (NGSS) for
science and the Common Core State Standards (CCSS) for math. The ultimate goals of
the standards are to prepare students to succeed in college and careers (Common Core
State Standards Initiative, 2010; NGSS Lead States, 2013). The fact that differences
were found between STEM graduates and non-STEM majors in Grade 6, highlights the
impact that early experiences may have on students and the need to implement
Grading Practices
Finally, it is important to recognize that the courses selected and the grading
practices for these courses directly affect the benchmarks. That is, different benchmarks
119
would have been found if different courses were selected. For example, a select group of
students in this study enrolled in advanced math courses beyond first-year calculus,
Spaces and Functions II. Because these classes were represented in total by only 30
students, they were not selected as typical first-year STEM courses. If these courses had
been included, the resulting Math STEM Readiness Benchmark would have been higher,
344 versus 331 found using the typical first-year STEM courses. This higher benchmark
would correspond to a National Percentile Rank (NPR) of 99 versus the NPR of 94 found
Different grading practices for the selected first-year STEM courses would have
also led to different benchmarks. The University of Iowa has recommended grade
distributions and while the science and math courses roughly followed these when all
courses were considered together, the grade distributions for individual courses varied
distributions if it was reasonably expected that the students represented atypical levels of
ability (CLAS recommended grade distributions, 2017). For instance, because honors
students are not an academically typical group, honors courses at the University of Iowa
are not required to follow the CLAS recommended grade distributions (Spisak, 2015).
Rather, instructors for these honors courses are advised that the grade distributions should
be heavily weighted toward the top grades. Because these are all STEM courses and
STEM students demonstrate higher achievement levels, it is reasonable to expect that the
instructors may have adjusted accordingly and assigned a larger percentage of As or Bs.
However, the differences found would not be fully explained through this scenario. For
120
example, while a larger proportion of STEM graduates enrolled in Principles of
40%, respectively). These findings highlight differences found for Iowa residents within
the University of Iowa. Grading practices may be more disparate when considering other
other universities proved difficult, as universities have stopped publicly reporting these
institutions are beyond the scope of this study, but would be interesting topics for future
research.
While grading practices affect the STEM Readiness Benchmarks, they also have
grades based on the percentage of students that will obtain each letter grade, similar to a
bell curve (Schneider & Hutt, 2014). For example, the recommended grade distribution
at the University of Iowa sets 50% of the grades for first year courses below a B (CLAS
recommended grade distributions, 2017). This practice assures that 50% of the students
will be deemed not college ready for these courses based on the STEM Readiness
Benchmarks. Furthermore, given these grading practices, roughly 50% of students who
enroll in first-year STEM courses will earn below a B. It may be that these grading
121
STEM programs, such as medical school. However, it is also worth considering that
these practices may eliminate students who actually would be successful in STEM
careers and could help to fill the need of more STEM graduates. First STEM courses
have a large impact as students often leave STEM majors because they are performing
worse in the classes than expected (Stinebrickner & Stinebrickner, 2014). These grading
practices may be enough to discourage “B” students, who were used to being “A”
students in high school, from pursing STEM careers. It is also possible that grading
practices may be partially responsible for the gender discrepancy in STEM graduates. As
Statistics, 2007; Huang et al., 2000) and confirmed in this study, females were more
likely to have higher GPAs in high school and college than males. Therefore, it is
discouraging to female students and may lead to the attrition of proportionally more
female students. The effects of grading practices on STEM attrition and as it relates to
Summary
Overall, the results from this study aim to assist students in making decisions
prior to college that will aid them in achieving their college goals. It is important for
students, educators, and policy makers to recognize that STEM graduates demonstrate
differences in achievement levels years before entering college. Recognizing this enables
educators to help students remain on-track for STEM readiness or offer appropriate
remedial education when students are not on-track to be ready for STEM coursework. In
addition, the STEM Readiness Benchmarks provide students and educators with
122
indicators of readiness for first-year STEM coursework in Science and Math. Knowing
one’s readiness for STEM coursework enables students to make informed decisions
prior to enrolling in college courses. This has important impacts on higher education,
because students who are prepared for STEM coursework are less likely to drop out of
Limitations
When considering the findings and implications of this study, it is also important
to consider the limitations. This study provided information related to the levels of
achievement levels were examined from Grade 6 to Grade 11 and matched to their
levels in earlier grades to determine when differences arise, but this was not possible due
to the nature of the available data. This study also compared STEM graduates with non-
STEM majors. Given concerns surrounding STEM attrition, it would have been
intriguing to compare these trends with students who began college as STEM majors, but
then switched to other majors. This information was also not available.
This study examined achievement levels on the IAs at the University of Iowa, a
large public university in the Midwest. The first-year STEM courses were found to be
similar to those selected from other four-year public institutions when setting ACT’s
STEM Readiness Benchmarks. While this provides evidence that these courses are
courses and grading practices are different at private institutions, including small
colleges. If so, this would also affect the level of academic achievement needed to be
123
successful at these institutions. Therefore, the generalizability of the results may be
Finally, there are some limitations inherent to the methods of this study. While
attempting to address them through changing the methods would change the very nature
and purpose of the study, it is important these limitations are noted. First, in this study,
STEM majors are defined as majors which require advanced science and math courses.
addition, this study only considered students enrolled at a 4-year institution and did not
include 2-year technical degrees or certifications. Next, the purpose of this study was to
compare STEM graduates with non-STEM majors. Given that students self-select to
attend college and self-select majors, selection bias is also a limitation of this study.
Suggestions for future areas of research have been mentioned throughout this
chapter and will be briefly reviewed now. It would be helpful to have further studies
examining the achievement levels needed for STEM success, but without the limitations
listed above. For example, examining achievement growth before Grade 6, examining
the trends for students who started as STEM majors but switched to non-STEM majors,
research could also include exploring the effects that different courses may have on the
found for different rates of growth from middle school to high school. Finally, future
124
growth by race/ethnicity or in accuracy of the STEM Readiness Benchmarks when
Conclusion
students are encouraged to pursue STEM majors it is especially pertinent that they have
accurate indicators of readiness for STEM college coursework. This study was the first to
date to examine achievement growth patterns for STEM majors. It was also the first study
graduates were found to have higher achievement levels in Grade 6 through Grade 11 in
Science, Math, and Reading. In addition, the STEM Readiness Benchmarks were higher
were highlighted throughout the study in an effort to better understand the gender
discrepancy in STEM graduates. The effects of grading practices were also discussed.
Findings from this study can be used to better understand the level of academic
students considering STEM careers. Future research on gender and grading practices
students are encouraged to consider STEM majors, educators are responsible for ensuring
students are adequately informed about the level of academic achievement needed to
succeed in STEM coursework. STEM Readiness Benchmarks are unique and valuable
125
APPENDIX A
126
Health and Human Physiology
Mathematics Calculus I
Nursing
Radiation Sciences
i
Grayed majors were not coded as STEM majors, because they did not require any of the core math or
science courses. Additionally, Computer Science I was not coded as a typical first-year STEM course,
because it was only required by two of the STEM majors.
127
APPENDIX B
Course Name
Math
Calculus for the Biological Sciences
Calculus I
Calculus II
Engineering Mathematics I: Single Variable Calculus
Engineering Mathematics II: Multivariable Calculus
Science
Principles of Biology I
Principles of Chemistry I
Introductory Physics I
Engineering Problem Solving I
128
APPENDIX C
Table C.1
Science
129
Table C.2
Math
130
Table C.3
Reading
131
REFERENCES
ACT, Inc. (2008). The forgotten middle: Ensuring that all students are on target for
college and career readiness before high school. Iowa City, IA: ACT, Inc.
ACT, Inc. (2014a). The condition of college & career readiness 2014 Iowa. Iowa City,
IA: ACT, Inc. Retrieved from http://www.act.org/newsroom/data/2014/states/
ACT, Inc. (2014b). The condition of STEM 2014. Iowa City, IA: ACT, Inc.
Retrieved from http://www.act.org/stemcondition/14/
ACT, Inc. (2015). The condition of college & career readiness 2015 Iowa. Iowa City,
IA: ACT, Inc. Retrieved from http://www.act.org/newsroom/data/2015/states/
Allen, J. (2013). Updating the ACT college readiness benchmarks. Iowa City, IA: ACT,
Inc.
Allen, J., & Sconing, J. (2005). Using ACT assessment scores to set benchmarks
for college readiness. Iowa City, IA: ACT Inc.
Alper, J. (1993). The pipeline is leaking women all the way along. Science, 260, 409-411.
doi: 10.1126/science.260.5106.409
Ansley, T., & Forsyth, R. (1983). Relationship of elementary and secondary school
achievement test scores to college performance. Educational and Psychological
Measurement, 43(4), 1103-1112. doi: 10.1177/001316448304300419
Board of Regents. (2009). Regent Admission Index (RAI) evaluation study report.
Des Moines, IA: Board of Regents, State of Iowa.
132
Board of Regents. (2013). Annual report of student retention and graduation rates.
Des Moines, IA: Board of Regents, State of Iowa.
Bollen, K. A. (1989). Structural equations with latent variables. New York: John
Wiley & Sons.
Burton, N., & Ramist, L. (2001). Predicting success in college: SAT studies of classes
graduating since 1980. (College Board Research Report No. 2001-2). New
York, NY: College Board.
Camara, W.J. (2013). Defining and measuring college and career readiness: A validation
framework. Educational Measurement: Issues and Practice 32(4), 16-27. doi:
10.1111/emip.12016
Carnevale, A., Smith, N., & Melton, M. (2011). STEM. Georgetown University Center
on Education and the Workforce. Washington, DC. Retrieved from
http://files.eric.ed.gov/fulltext/ED525297.pdf
Castellano, K., & Ho, A. (2013). A practitioner’s guide to growth models. Council
of Chief State School Officers.
Chang, M., Sharkness, J., Hurtado, S., & Newman, C. (2014). What matters in college for
retaining aspiring scientists and engineers from underrepresented racial groups.
Journal of Research in Science Teaching, 51(5), 555-580. doi: 10.1002/tea.21146
Chen, X. (2013). STEM Attrition: College Students’ Paths Into and Out of STEM Fields
(NCES 2014-001). National Center for Education Statistics, Institute of Education
Sciences, U.S. Department of Education. Washington, DC.
Choi, K., & Goldschmidt, P. (2012). A multilevel latent growth curve approach to
predicting student proficiency. Asia Pacific Education Review, 13(2), 199-
208. doi: 10.1007/s12564-011-9191-8
Chou, C.-P., Yang, D., Pentz, M. A., & Hser, Y.-I. (2004). Piecewise growth curve
modeling approach for longitudinal prevention study. Computational Statistics &
Data Analysis, 46, 213-225. doi: 10.1016/S0167-9473(03)00149-X
Choy, S. (2002). Access and persistence: Findings from 10 years of longitudinal research
on students. American Council on Education. Washington, D.C.
133
Common Core State Standards Initiative. (2010). Retrieved from
http://www.corestandards.org
Clinedinst, M., Hurley, S., & Hawkins, D. (2011). 2011 state of college admission.
Arlington, VA: National Association for College Admission Counseling.
Davison, M. L., Jew, G. B., & Davenport, E. C., Jr. (2014). Patterns of SAT scores,
choice of STEM major, and gender. Measurement and Evaluation in
Counseling and Development, 47(2), 118-126.
Fina, A. D. (2014). Growth and the college readiness of Iowa students: A longitudinal
study linking growth to college outcomes. (Doctoral dissertation). Iowa City, IA:
University of Iowa.
Forsyth, R.A., Ansley, T.N., Feldt, L.S., & Alnot, S.D. (2001). Iowa Tests of Educational
Development. Itasca, IL: Riverside Publishing.
Forsyth, R.A., Ansley, T.N., Feldt, L.S., & Alnot, S.D. (2003). Iowa Tests of Educational
Development: Norms and score conversions. Itasca, IL: Riverside Publishing.
Furgol, K., Fina, A.D., & Welch, C. (2011). Establishing validity evidence to
assess college readiness through a vertical scale. Paper presented at the
American Educational Research Association, New Orleans, LA.
Griffith, A. (2010). Persistence of women and minorities in STEM field majors: Is it the
school that matters? Economics of Education Review, 29(6), 911–922.
doi: 10.1016/j.econedurev.2010.06.010
Halpern, D., Benbow, C., Geary, D., Gur, R., Hyde, J., & Gernsbacher, M. (2007). The
science of sex differences in science and mathematics. Psychological Science
in the Public Interest, 8(1), 1-51. doi: 10.1111/j.1529-1006.2007.00032.x
Heidel, J., Ali, H., Corbett, B., Liu, J., Morrison, B., O'Connor, M., . . . Ryan, C.
(2011). Increasing the number of homegrown STEM majors: What works and
what doesn't. Science Educator, 20(1), 49-54.
Heilman, M., Wallen, A., Fuchs, D., & Tamkins, M. (2004). Penalties for success:
Reactions to women who succeed at male gender-typed tasks. Journal of Applied
Psychology, 89(3), 416-427.
134
Heilbronner, N. N. (2009). Pathways in STEM: Factors affecting the retention and
attrition of talented men and women from the STEM pipeline. (Doctoral
dissertation). Retrieved from http://ebscohost.com/
Heilbronner, N. N. (2013). The STEM pathway for women: What has changed? Gifted
Child Quarterly, 57(1), 39-55.
Hoover, H.D., Dunbar, S.B., & Frisbie, D.A. (2001). Iowa Tests of Basic Skills. Itasca,
IL: Riverside Publishing.
Hoover, H.D., Dunbar, S.B., & Frisbie, D.A. (2003). Iowa Tests of Basic Skills: Norms
and score conversions. Itasca, IL: Riverside Publishing.
Hox, J. J., & Stoel, R. D. (2005). Multilevel and SEM approaches to growth curve
modelling. In B. S. Everitt & D. C. Howell (Eds.), Encyclopedia of statistics
in behavioral science (pp. 1296 –1305). Chichester, England: Wiley.
Hu, L., & Bentler, P. (1999). Cutoff criteria for fit indexes in covariance structure
analysis: Conventional criteria versus new alternatives. Structural
Equation Modeling: A Multidisciplinary Journal, 6(1), 1-55. doi:
10.1080/10705519909540118
Huang, G., Taddese, N., & Walter, E. (2000). Entry and persistence of women and
minorities in college science and engineering education. Education
Statistics Quarterly, 2(3), 59-60.
Hyde, J., Lindberg, S., Linn, M., Ellis, A., & Williams, C. (2008). Gender
similarities characterize math performance. Science, 321, 494-495.
Jones, L., Mullis, I., Raizen, S., Weiss, I., & Weston, E. (1992). The 1990 science
report card: NAEP’s assessment of fourth, eighth, and twelfth graders.
Princeton, NJ: Educational Testing Services.
Kass, R. E., & Raftery, A. E. (1995). Bayes Factors. Journal of American Statistical
Association, 90(730), 773-795.
Kline, R.B. (2011). Principles and practice of structural equation modeling (3rd ed.).
New York, NY: Guilford Press.
Kohli, N., Harring, J. R., & Hancock, G. R. (2013). Piecewise linear–linear latent
growth mixture models with unknown knots. Educational and Psychological
Measurement, 73(6), 935-955. doi: 10.1177/0013164413496812
135
LeBeau, B., Harwell, M., Monson, D., Dupuis, D., Medhanie, A., & Post, T. (2012).
Student and high-school characteristics related to completing a science,
technology, engineering, or mathematics (STEM) major in college. Research in
Science & Technological Education, 30(1), 17-28. Doi:
10.1080/02635143.2012.659178
Li, M., Shavelson, R., Kupermintz, H., & Ruiz-Primo, M. (2002). On the relationship
between mathematics and science achievement in the United States. In D.
Robitaille and A. Neaton (Eds.), Secondary analysis of the TIMSS data (pp. 233-
249). Netherlands: Kluwer Academic Publishers.
Lindquist, E.F. (1942). A new testing service: Fall testing programs for Iowa high
schools. Iowa City, IA: University of Iowa.
Lloyd, P., & Eckhardt, R. (2010). Strategies for improving retention of community
college students in the sciences. Science Educator, 19(1), 33-41. Retrieved
from http://www.eric.ed.gov
Loyd, B.H., Forsyth, R.A., & Hoover, H.D. (1980). Relationship of elementary and
secondary school achievement test scores to later academic success. Educational
and Psychological Measurement, 40(4), 1117-1124. doi:
10.1177/001316448004000441
Lubinski, D., & Benbow, C. (2006). Study of mathematically precocious youth after 35
years: Uncovering antecedents for the development of math-science expertise.
Perspectives on Psychological Science 1(4), 316-345. doi: 10.1111/j. 1745-
6916.2006.00019.x
Mattern, K., Radunzel, J., & Westrick, P. (2015). Development of STEM readiness
benchmarks to assist educational and career decision making. (ACT
Research Report Series No. 2015-3). Iowa City, IA: ACT, Inc. Retrieved from
http://www.act.org/research/researchers/reports/pdf/ACT_RR2015-3.pdf
Miller, D., & Wai, J. (2015). The bachelor’s to Ph.D. STEM pipeline no longer leaks
more women than men: A 30-year analysis. Frontiers in Psychology, 6(37), 1-10.
doi: 10.3389/fpsyg.2015.00037
Moore, G., Slate, J., Edmonson, S., Combs, J., Bustamante, R., & Onwuegbuzie, A.
(2010). High school students and their lack of preparedness for college: A
study statewide. Education and Urban Society, 42(7), 817-838. doi:
10.1177/0013124510379619
136
Muller, P., Stage, F., & Kinzie, J. (2001). Science achievement growth trajectories:
Understanding factors related to gender and racial-ethnic differences in precollege
science achievement. American Educational Research Journal, 38, 981–1012.
Muthén, B.O. (1997). Latent variable growth modeling with multilevel data. In M.
Berkane (Ed.), Latent variable modeling and applications to causality (pp. 141-
161). New York, NY: Springer.
Muthén, B.O., & Khoo, S. (1998). Longitudinal studies of achievement growth using
latent variable modeling. Learning and individual differences, 10(2), 73 -101.
doi: 10.1016/S1041-6080(99)80135-6
Muthén, L.K., & Muthén, B.O. (1998-2012). Mplus user’s guide (7th ed.). Los
Angeles, CA: Muthén & Muthén.
National Center for Education Statistics. (2011). The nation’s report card: Mathematics
2011. Washington, DC: U.S. Department of Education, Institute of Education
Sciences.
NGSS Lead States (2010). Next Generation Science Standards: For States. Retrieved
from http://www.nextgenscience.org
No Child Left Behind Act of 2001, Pub. L. No. 107-110, § 115, Stat. 1425 (2002).
Nylund, K.L., Asparouhov, T., & Muthén, B.O. (2007). Deciding on the number of
classes in latent class analysis and growth mixture modeling: A Monte
Carlo simulation study. Structural Equation Modeling, 14(4), 535-569. doi:
10.1080/10705510701575396
Ost, B. (2010). The role of peers and grades in determining major persistence in
sciences. Economics of Education Review, 29(6): 923−934.
Phillip, N., Brennan, T., & Meleties, P. (2005). Increasing enrollment in chemistry
courses and programs. The Chemical Educator, 10(3), 227-230.
Preacher, K. J., Wichman, A. L., MacCallum, R. C., & Briggs, N. E. (2008). Chapter
1: Introduction. In K. J. Preacher, A. L. Wichman, R. C. MacCallum, & N. E.
Briggs (Eds.). Latent growth curve modeling (pp. 1-21). Thousand Oaks, CA:
SAGE Publications, Inc. doi: http://dx.doi.org/10.4135/9781412984737
Qualls, A.L., & Ansley, T.N. (1995). The predictive relationship of ITBS and ITED to
measures of academic success. Educational and Psychological Measurement,
55(3), 485-498. doi: 10.1177/0013164495055003016
137
Quinn, D., & Cooc, N. (2015). Science achievement gaps by gender and race/ethnicity
in elementary and middle school: Trends and predictors. Educational
Researcher, 44(6), 336-346.
Radunzel, J., & Nobel, J. (2012). Predicting long-term college success through degree
completion using ACT composite score, ACT benchmarks, and high school grade
point average. (ACT Research Report No. 2012-5). Iowa City, IA: ACT, Inc.
Rask, K. (2010). Attrition in STEM fields at a liberal arts college: The importance of
grades and pre-collegiate preferences. Economics of Education Review,
29(6): 892−900.
Riegle-Crumb, C., King, B., Grodsky, E., & Muller, C. (2012). The more things change,
the more they stay the same? Prior achievement fails to explain gender inequality
in entry into STEM college majors over time. American Educational Research
Journal, 49(6), 1048-1073.
SAS Institute Inc. (2011). Base SAS® 9.3 procedures guide. Cary, NC: SAS Institute
Inc.
Sawyer, R. (2010). Usefulness of high school average and ACT scores in making college
admission decisions. (ACT Research Report No. 2010-2). Iowa City, IA: ACT,
Inc.
Schmitt, N., Keeney, J., Oswald, F., Pleskac, T., Billington, A., Sinha, R., & Zorie,
M. (2009). Prediction of 4-year college student performance using cognitive
and noncognitive predictors and the impact on demographic status of admitted
students. Journal of Applied Psychology, 94(6), 1479-1497. doi:
10.1037/a0016810
Schneider, J., & Hutt, E. (2014). Making the grade: A history of the A-F marking scheme.
Journal of Curriculum Studies, 46(2), 201-224. doi:
10.1080/00220272.2013.790480
Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6(2),
461-464.
Shapiro, C. A., & Sax, L. J. (2011). Major selection and persistence for women in
STEM. New Directions for Institutional Research, 152, 5-18.
138
Sharpe, D. (2015). Your chi-square test is statistically significant: Now what? Practical
Assessment, Research and Evaluation, 20(8), 1-10. Retrieved from
http://pareonline.net/
Stinebrickner, R., Stinebrickner, T. (2014). A major in science? Initial beliefs and final
outcomes for college major and dropout. The Review of Economic Studies, 81(1),
426-472.
Stoel, R., van den Wittenboer, G., & Hox, J. (2004). Including time-invariant
covariates in the latent growth curve model. Structural Equation Modeling: A
Multidiciplinary Journal, 11(2), 155-167. doi: 10.1207/s15328007sem1102_1
Tai, R. H., Liu, C. Q., Maltese, A. V., & Fan, X. (2006). Planning early for careers
in science. Science, 312, 1143-1144.
University of Iowa. (2008). The University of Iowa census count fall 2008 (end of day
10). Retrieved from: http://ir.uiowa.edu/registrar_census/18/
U.S. Department of Commerce. (2011a). STEM: Good jobs now and for the future.
Economics and Statistics Administration Issue Brief, 3(11), 1-10. Retrieved from
http://www.esa.doc.gov
U.S. Department of Education. (2010). Race to the Top Assessment program: Executive
summary. Washington, DC: U.S. Department of Education.
139
U.S. Department of Education. National Center for Educational Statistics. (2007). The
nation’s report card: America’s high school graduates: Results from the 2005
NAEP high school transcript study, by C. Shettle, S. Roey, J. Mordica, R. Perkins,
C. Nord, J. Teodorovic, J. Brown, M. Lyons, C. Averett, & D. Kastberg. (NCES
2007-467). Washington, DC: Government Printing Office.
U.S. Department of Labor. (2007). The STEM workforce challenge: The role of the
public workforce system in a national solution for a competitive science,
technology, engineering, and mathematics (STEM) workforce. Washington,
DC: Author. http://digitalcommons.ilr.cornell.edu/key_workplace/637
Wang, X. (2013). Why students choose STEM majors: Motivation, high school
learning, and postsecondary context of support. American Educational Research
Journal, 50(5), 1081-1121.
Wang, M., Chen, K., & Welch, C. (2012). Evaluating college readiness for English
Language Learners and Hispanic and Asian students. Iowa City, IA: Iowa
Testing Programs.
Weinberger, C. (2005). Is the science and engineering workforce drawn from the
far upper tail of the math ability distribution? Working paper. Retrieved
from http://econ.ucsb.edu/~weinberg/uppertail.pdf
The White House, Office of the Press Secretary. (2009). President Obama launches
“Educate to Innovate” campaign in excellence in science, technology,
engineering, & math (STEM) education [Press release]. Retrieved from
https://www.whitehouse.gov/the-press-office/president-obama-launches-educate-
innovate-campaign-excellence-science-technology-en
The White House, Office of the Press Secretary. (2011). Remarks by the President in
State of Union Address [Press release]. Retrieved from
https://www.whitehouse.gov/the-press-office/2011/01/25/remarks-president-state-
union-address
The White House, Office of the Press Secretary. (2015). FACT SHEET: President
Obama announces over $240 million in new STEM commitments at the 2015
White House science fair [Press release]. Retrieved from
https://www.whitehouse.gov/the-press-office/2015/03/23/fact-sheet-president-
obama-announces-over-240-million-new-stem-commitmen
140
Williamson, G.L., Appelbaum, M., & Epanchin, A. (1991). Longitudinal Analyses of
Academic Achievement. Journal of Educational Measurement, 28(1), 61-76. doi:
10.2307/1434685
Willingham, W., Lewis, C., Morgan, R., & Ramist, L. (1990). Predicting college grades:
An analysis of institutional trends over two decades. Princeton, NJ: Educational
Testing Service.
Wyatt, J., Remigio, M., & Camara, W. (2013). SAT subject area readiness indicators:
Reading, writing, and STEM. College Board Research Report 2013-4. Retrieved
from http://research.collegeboard.org/sites/default/files/publications/2013/1/
researchnote-2013-1-sat-subject-readiness-indicators_0.pdf
Zvoch, K., & Stevens, J.J. (2006). Longitudinal effects of school context and practice
on middle school mathematics achievement. The Journal of Educational
Research, 99(6), 347-357. doi: 10.3200/JOER.99.6.347-357
141