Abstract
In this study, we conducted binary logistic regression on survey data collected from 244 past participants of a Talent Search program who attended regular high schools but supplemented their regular high school education with enriched or accelerated math and science learning activities. The participants completed an online survey 4 to 6 years after high school. This study examined how their demographics, high school experiences, and timing of and reasons for pursuing a science, technology, engineering, and mathematics (STEM) pathway related to the probability of earning STEM college degrees. This study revealed two factors that were positively and significantly associated with the outcome of earning STEM college degrees: Asian or White ethnicity and students’ personal interest in STEM. Findings suggest that students’ success in earning STEM degrees may not be fully attributable to their high achievements or abilities, and that their experiences in the Talent Search and supplemental outside-of-school gifted programs helped students intensify their interests in STEM.
Building a proficient science, technology, engineering, and mathematics (STEM) workforce has become an issue central to U.S. economic growth and global competitiveness. The demand for a skilled STEM workforce has been increasing and will continue to grow. The Bureau of Labor Statistics (BLS) projected that the total employment in science and engineering occupations will increase 18.7% between 2010 and 2020, compared with an increase of 14.3% in all occupations combined (National Science Board [NSB], 2014). This poses a looming challenge to increase the domestic supply of STEM talents. NSB (2010) reported that although the total number of bachelor’s degrees awarded annually in the United States has nearly tripled over the last 40 years, the relative percentage of students earning STEM degrees in 2006 was approximately equal to or even lower than that observed in the previous 40 years. Bachelor’s degrees in science and engineering have consistently accounted for approximately only one third of all bachelor’s degrees awarded between 2001 and 2010 (NSB, 2014).
To meet the rising challenge, efforts have been made to reinvigorate an interest in STEM education and make it a national priority. In its 2012 Executive Report to the President, the President’s Council of Advisors on Science and Technology (PCAST) identified seven high-priority recommendations for STEM education. One of these recommendations urged the Federal Government to “promote the creation of at least 200 new highly-STEM-focused high schools and 800 STEM-focused elementary and middle schools over the next decade, including many serving minority and high-poverty communities” (PCAST, 2012, p. х). Another advised to “develop a coordinated initiative . . . to support the development of a wide range of high-quality STEM-based afterschool and extended-day activities (such as STEM contests, fabrication laboratories, summer and afterschool programs, and similar activities)” (p. iх). These recommendations underscore a pressing need for alternative learning opportunities inside and outside of formal classrooms to develop STEM talents.
In the current environment of a reinvigorated national focus on excellence in STEM, the NSB (2010) set forth a core mission for the field of gifted education: identify and develop the most talented and motivated students into the next generation of STEM innovators who would then become the leading world STEM professionals and creators of significant breakthroughs and advances in science and technology. Focusing on this mission, the NSB set forth three keystone recommendations: The first recommendation was to provide opportunities for excellence. To achieve this goal, the NSB presented eight policy actions, including “Increase access to and quality of college-level, dual enrollment, and other accelerated coursework, as well as high-quality enrichment programs” (Policy Action B, p. 2), and “Increase access to create a national database of formal and informal education opportunities for highly talented students, and publicize and promote such opportunities nationally to parents, education professionals, and content and resource providers” (Policy Action H, p. 3). Taken together, these recommendations and policy actions call for promoting the use of gifted programs and outside-of-school learning to improve STEM education for all students, especially gifted ones.
Previous Research on Factors Important to STEM Pathways
A substantial body of research has studied factors that contribute to students’ success along the STEM pathways (Harackiewicz, Rozek, Hulleman, & Hyde, 2012; Hidi & Harackiewicz, 2000; Maltese & Tai, 2010). Based upon a theory that pursuing a STEM major is a longitudinal process that builds during secondary education and continues into postsecondary stages, the bulk of extant research has focused on the impact of high-school-based factors on students’ decisions, plans, persistence, and completion of STEM degrees. Common high-school-based factors that are studied include student demographics, aspects of their high school experience, and the timing of interest in and reasons for pursuing a STEM pathway.
Demographic Factors
Much research has examined the effect of demographic variables such as gender, ethnicity, family socioeconomic status (SES), parent education, and language on a student’s likelihood of earning a degree in a STEM-related area (e.g., Dabney, Chakraverty, & Tai, 2013; Tai, Liu, Maltese, & Fan, 2006). Exemplary findings include that male high school students from a high-SES family are more likely to choose science majors than female students and/or male students from disadvantaged family backgrounds (see Ware & Lee, 1988). Female students, even those who complete advanced-level science and mathematics courses in high school, are less likely to earn STEM degrees than male students (Tyson, Lee, Borman, & Hanson, 2007). Furthermore, research shows that Asian American students are more likely to earn STEM degrees than White students (Crisp, Nora, & Taggart, 2009).
High School Experience
Much research has also centered on the relationship between students’ high school experiences and success in STEM pathways. This line of research focuses on aspects of the high school experience, including course selection, course sequences, school environment, and students’ motivation, perceptions, and attitudes toward STEM subjects and careers. For example, Almarode et al. (2014) conducted a national, retrospective, online survey study on the impact of specialized selective STEM high schools. The central goal of this study was to examine whether and how students’ experiences in selective specialized STEM high schools were associated with the outcome of earning STEM-related college degrees approximately 4 to 6 years after high school graduation. These students were compared with equally competent students who enrolled in regular high schools but supplemented their school learning with outside-of-school enrichment and accelerated courses in science and math through regional Talent Search programs. They found that specialized STEM high schools and Talent Search programs, which offered a wide array of STEM courses and exceptional teachers, served equally well as incubators of STEM-talented adolescents (also see Subotnik, Tai, Almarode, & Crowe, 2013).
A number of studies have found that planning to major in STEM or earn STEM degrees is associated with taking more high school science and mathematics courses (e.g., Tyson et al., 2007). Research also shows that belief in one’s ability to achieve in STEM areas, as well as the quality of students’ academic experience in high school (e.g., level of challenge, opportunities for hands-on learning, and preparation for careers), contributes to gifted students’ declaration of STEM college majors (see Heilbronner, 2011).
Another critical feature of students’ educational environments that can influence their intrinsic motivation to pursue STEM studies and/or careers is their sense of belonging (Good, 2012). Sense of belonging refers to individuals’ need to feel connected personally and interpersonally to a domain of achievement such as math, engineering, or science. Negative stereotypes (e.g., females are not as talented in STEM) and fixed mindsets can adversely affect sense of belonging (Good, 2012), while a positive feeling of belonging can facilitate long-term commitment and perseverance during difficult times and ultimate achievement. Teachers can cultivate a sense of belonging within their classrooms by setting high expectations and working to ensure the success of all their students (Farrington et al., 2012).
Timing of and Reasons for STEM Pathways
Another area of relevant research centers on the timing of and reasons for becoming interested in or deciding to enter STEM fields. For instance, Maltese and Tai (2011) analyzed longitudinal data from the National Education Longitudinal Study of 1988 (NCES, 1988) to study how experiences in science and math classes, attitudes, and their performance influenced students’ decisions to remain in or leave the STEM pathways from adolescence through early adulthood. They found that most students who majored in STEM made that choice during high school, and they did so because of growing interest in math and science. Similarly, analyzing the Education Longitudinal Study of 2002 (NCES, 2002), Wang (2013) found that choosing a STEM major was directly influenced by students’ intent to major in STEM as well as their high school math achievement.
The previously mentioned study of the impact of specialized STEM high schools (see Almarode et al., 2014; Subotnik et al., 2013; from here on called the “national STEM high school study”) investigated whether there was a significant difference in the odds of earning STEM college degrees among the different types of specialized STEM high schools (e.g., school-within-schools, residential, half-time, or full-time commuter schools). The authors conducted some preliminary analyses to explore whether factors such as self-efficacy and stability of STEM interests were associated with the outcomes of earning STEM college degrees. They found that students’ perceptions of their intellectual capacity to be scientists, mathematicians, and/or engineers as juniors or seniors in high school were associated with a higher likelihood that they will earn a STEM degree in college. Moreover, students whose interests in STEM intensified or grew were 4.448 times more likely to earn STEM degrees than those whose interests were not stabilized (see Almarode et al., 2014).
Purpose of the Current Study
This study utilized part of the data collected from the online survey in the national STEM high school study—specifically the data on 261 Talent Search participants, a subsample of the study. The current study explored factors that might be significantly associated with the outcomes of earning STEM college degrees but were not fully investigated in the national STEM high school study. Specifically, the current study addressed three key research questions:
Method
Participants
The participants of the current study were among the 603 comparison subjects in the national STEM high school study. Specifically, the subjects were 261 past participants from the Northwestern University Midwest Academic Talent Search (NUMATS) program. The remaining 342 of the comparison subjects in the STEM high school study were past participants in a Talent Search program in the northeastern part of the United States, whose data were not available to us.
Talent Search programs have been in existence for more than 30 years and are currently operated at several U.S. universities, including Northwestern University, Duke University, and Johns Hopkins University. Talent Search programs are guided by the philosophy that appropriate assessment is a critical first step to creating an appropriate match between students’ abilities and levels of achievement and academic programming. The core of the talent search model is domain-specific and above-grade-level testing. Currently, this is embodied in the Talent Search programs’ uses of tests, such as the American College Test (ACT) and Scholastic Aptitude Test (SAT), to assess the reasoning abilities of middle school students (Corwith & Olszewski-Kubilius, 2012). In addition, Talent Search programs provide enriched, accelerated, and supplemental academic services to identified students on weekends, summers, or online. Talent Search is a well-researched model of identification and programming for academically gifted learners (Olszewski-Kubilius, 2004, 2014, 2015; Olszewski-Kubilius & Thomson, 2014).
The Talent Search program at Northwestern University, NUMATS, utilizes above-grade-level tests for students in Grades 3 to 12. Participants in the NUMATS program need to meet at least one of the following criteria: (a) qualify for their school’s gifted/talented programs; (b) have been nominated by a teacher or parent for advanced aptitude in verbal or mathematical reasoning, demonstrated by a consistently high level of performance in demanding coursework; and (c) meet grade-level assessment criteria in either reading or math on a nationally normed or state achievement test. For example, third through ninth graders have to score in at least 90th percentile or above in verbal, reading, math, or composite on nationally normed tests or state achievement tests to be eligible for NUMATS in the 2014–2015 school year.
Researchers of the national STEM high school study screened and selected 1,442 potential participants from 33,961 past participants of the NUMATS with three criteria: (a) scored high enough in above-grade-level standardized tests, such as the ACT or the SAT, to qualify for enrichment and accelerated math and science courses through the Center for Talent Development (CTD) of Northwestern University; (b) demonstrated interest in STEM learning by taking enriched or accelerated math and science courses through the CTD; and (c) did not attend a selective specialized STEM high school. In the end, 261 (of 1,442) of the NUMATS potential participants completed an extensive online survey and were included as part of the comparison group in the national STEM high school study. These students participated in the NUMATS between the years 2004 and 2007. We initially intended to include all 261 students. However, we eventually excluded 17 respondents who self-reported that they attended specialized STEM high schools. The remaining 244 were graduates of non-STEM high schools and therefore, became the participants of the current study.
The Survey
As mentioned above, the data in the current study were collected from an online survey in the national STEM high school study. This survey consisted of 118 question items derived from research literature on instruments that assessed motivation, attitudes, and career development of individuals in science. To ensure the validity and reliability of this survey, the authors first piloted survey questions with graduates of the Illinois Mathematics and Science Academy (IMSA) and a group of Talent Search participants. We revised the survey based on the results of the first pilot, and further tested with graduates of three additional specialized high schools and a group of Talent Search participants. The final survey incorporated some revisions and improvement from the second pilot. The majority of the questions inquired about participants’ high school experience, especially learning experiences in science, math, and/or technology subjects. For example, the survey inquired the kind of high school students attended, the most important reasons for their decision to pursue advanced STEM subjects in high school, the nature of their laboratory-related learning in high school, and the sources and timing of their initial interests in STEM. The survey also inquired of participants to indicate their undergraduate major or concentration from a provided list of 24 categories (e.g., biological sciences, chemistry, engineering; see Table 1 for the full list).
Key Variables, Survey Questions, Original Responses, and Coded Categories.
Note. The survey questions and responses were slightly revised from the original survey in Almarode et al.’s (2014) and Subotnik, Tai, Almarode, and Crowe’s (2013) studies. For example, the current study used the term STEM, whereas the original survey used the term SMT as an abbreviation for science, mathematics, and technology. STEM = science, technology, engineering, and mathematics.
“Variable type” (i.e., one-off or scale variable) denotes a basic feature of the coded variables. A one-off variable means that such a variable was associated with a survey question to which the respondents could only select from one of the two given options, such as “White” or “No.” The two subcategories of a one-off variable, such as “White” and “No,” only indicate a qualitative difference. For a scale variable, its related subcategories, such as “Interests sustained or grew” and “Interests Lost,” often indicate a quantitative or degree difference (e.g., more or less) between the subcategories.
Data Collection
In the national STEM high school study, the survey was conducted using e-mail solicitations. Contact information of potential participants was obtained through a combination of school records from 25 selective specialized STEM high schools, program records of the two Talent Search programs (including NUMATS), and alumni association lists of partnering schools and programs. The survey was conducted between 2010 and 2013; about 4 to 6 years after these students had graduated from high school. This 4- to 6-year window is an appropriate time to examine career-related outcomes. Of the 9,461 e-mail solicitations, 4,113 (43.5%) completed the online survey (see Almarode et al., 2014, and Subotnik et al., 2013, for more detailed information regarding data collection).
Coding
Data coding involved assigning variable names to represent the information collected from the original survey questions. It also involved organizing and grouping the participants’ responses to some survey questions. As Table 1 shows, the number of response options for a given survey question varied from 2 to 10. Including such detailed or narrowly designated responses into the analyses might lead to a reduced statistical power, given a moderate sample size of the current study. Furthermore, results from such analyses could be difficult to interpret. Therefore, it was necessary and reasonable to recode and regroup the original survey responses into relatively broader subcategories. In sum, coding and grouping of the responses built a sound data structure for analyses and for the interpretation of the results.
Table 1 presents key variables of the current study, their corresponding original survey questions and associated response options, and the coded response categories for each variable. It should be noted that all coded variables were dichotomous, having two subcategories, such as “Yes” or “No” and “Male” or “Female.” Such two-level dichotomous variables correspond well with survey questions that had two predesignated response options. For some variables, however, the original survey responses had more than two levels. For example, for the variable “STEM interest,” there were four predesignated response options:
I permanently lost interest at some point during my time as a student.
I temporarily lost interest at some point during my time as a student.
I never lost interest.
My interest intensified or grew over time.
We coded the original response options 1 and 2 into a subcategory called “Interests lost,” and the responses 3 and 4 into a second subcategory called “Interests sustained or grew.” We applied similar coding strategies to several other variables that had more than two levels of response options, including Parent education, Hands-on, Field trip, Valuable, Belonging, Select major, Pressure, Passion, and Advanced study.
It is important to note that the decision to code original responses into broader subcategories for any given variable was made with practical and conceptual considerations. Coding also took into account the frequencies of original responses. For example, the variable “Select major” corresponded with the survey question: “When did you select the major or concentration with which you would eventually graduate?” There were seven predesignated responses for this question:
Elementary school or earlier
Middle school
High school
During my freshman of college
During my sophomore year of college
During my junior year of college
During my senior year of college
The number of respective participants for each of the responses was 5, 3, 46, 58, 57, 28, and 4. We combined the first three response options into a subcategory “Select major early.” We then combined response options 4 to 7 into the “Select major late” subcategory. We chose not to group the first two response options into a separate subcategory because there were only eight respondents in total for the two response options. In summary, the dichotomous nature of the coded variables was based on the practical and conceptual considerations as well as the frequency of the original responses.
Outcome Variable
The outcome variable in the current study was the acquisition of a STEM college degree. As Table 1 shows, it consisted of two dichotomous subcategories: STEM-related degrees and non-STEM-related degrees. Biological sciences, chemistry, engineering, mathematics/statistics, physics, behavioral sciences, computer science, and/or environmental science were coded as STEM-related college majors or concentrations. Sociology, religious studies, psychology, philosophy, history, political science, foreign language, fine arts, economics, English, business, communications, education, prelaw, health professions (e.g., pre-Med, nursing), and others were coded as non-STEM-related majors. This categorization of majors was in line with the practice found in the national STEM high school study.
Predictor Variables
We purposefully selected three groups of predictor variables: demographic, high school experience, and variables regarding timing of and reasons for STEM pathways (see Table 1). We selected these predictor variables based on two considerations: First, similar variables have been shown to be influential on a student’s decision or eventual completion of a STEM degree in research literature (e.g., Almarode et al., 2014; Heilbronner, 2011; Maltese & Tai, 2010; Tyson et al., 2007; Wang, 2013). Second, the total number of variables in each analysis model ought to not exceed 24, the maximum number of variables allowed in a binary logistic model given a sample size of 244. Researchers have recommended a minimum ratio of 10 to 1 for the number of participants and variables, with a minimum sample size of 100 or 50 for multivariate statistical analysis (Marascuilo & Levin, 1983; Tabachnick & Fidell, 2001).
Demographic variables
We initially selected nine demographic variables in preliminary analyses. They were Gender, White, Asian, Black, Hispanic, Parent job (i.e., whether parents/guardians had a STEM-related job), Parent education (i.e., whether the parents/guardian had a doctoral-level education), Born in the United States (i.e., whether the participants were born in the United States), and Language (i.e., whether English was the primary language spoken at home).
High school experience variables
We selected eight variables to indicate three aspects of high school experiences. Two variables (i.e., Public and Independent school) indicated the type of high school students graduated from (i.e., whether they attended a public or independent/parochial high school). Three variables (i.e., Hands-on, Research, and Field trip) related to participants’ high school experience concerning STEM-related learning, specifically how often students engage in hands-on experience at their high school, whether they participated in or performed original scientific research in an active research laboratory while in high school, and how often they engaged in STEM-related field trips at their high school. The remaining three variables (i.e., Prepared, Valuable, and Belonging) conveyed information on students’ perceptions of their high school experience in terms of how well prepared they felt for college-level work compared with their college classmates, whether they found their high school preparation in STEM subjects “valuable” to their chosen field of study, and whether they had strong feelings of belonging during their high school years.
Timing of and reasons for STEM pathways
Four variables (i.e., Select major, Pressure, Passion, and STEM interest) related to the timing of and reasons for STEM pathways. The variable Select major indicated whether the participants selected the major or concentration with which they eventually graduated “early” (i.e., high school or earlier) or “late” (i.e., in college). The variable Pressure indicated whether the participants felt pressure any time from family members, teachers, or other sources to pursue a STEM major or career. The variable Passion indicated whether participants’ passion for STEM was the most important in their decision to pursue advanced STEM subjects in high school, as opposed to other reasons, such as seeking prestige and recognition or getting a scholarship to attend a university. The variable STEM interests captured whether the participants permanently or temporarily lost interests in STEM at some point during their time as students or their interests in STEM sustained or grew over time.
College and beyond variables
We included three variables called Change major, Advanced study, and Career satisfaction. These variables corresponded to three survey questions concerning issues during or after college:
While in college, did you change your major or concentration?
At what level are you currently enrolled in school?
Are you satisfied with the direction you have pursued thus far in your professional life?
With the Change major variable, we examined whether participants who changed majors in college were more or less likely to earn STEM degrees. With the other two variables, we studied whether earning a STEM degree significantly predicted the odds of the participants engaging in advanced studies or feeling satisfied with their career life.
Data Analysis
Descriptive analyses
We conducted descriptive analyses on all 24 key variables (see Table 2). For each variable, we reported the frequency and percentage of its two subcategories. Although the total number of participants was 244, the number of respondents for different variables ranged from 177 (for the Career satisfaction variable) to 244, due to missing values. For example, 223 participants reported their gender identity. Of them, 101 (45%) were males, and 122 (55%) were females.
Descriptive Information of Key Variables.
Note. STEM = science, technology, engineering, and mathematics.
The total number of participants was 244. However, the number of respondents for different variables ranged from 177 to 244 due to missing values.
Sixty respondents reported that their interests in STEM remained, whereas 60 reported that their interests intensified or grew over time.
Table 3 presents five survey questions for which the response options were not exclusive in nature. For these survey questions, for example, the respondents were given the choice to “select all that apply.” Such data were not appropriate for advanced statistical analyses, which require data independence. In addition, Table 3 includes a follow-up question: “When did this loss of interest occur?” This question was answered only by those who answered “losing their interests in STEM” to the prior question.
Descriptive Information of Selected Survey Questions and Responses.
Note. STEM = science, technology, engineering, and mathematics.
The sum of percentages of listed responses under each related survey question did not equal 100. This is because the percentage of each response was calculated with the total number of respondents for that particular response as the denominator.
Each response under its related survey question ought to be viewed individually. This is because the survey was conducted online in a way in which each predesigned response was presented to the respondents one at a time with two options. For example, for the question “To the best of your recollection, which years of schooling provided the experiences that were MOST important in determining your choice of major or concentration? Please select all that apply,” the first response appeared with two options: “Elementary school or earlier” and “Null.” The respondents were allowed to choose one of the options. In this example, 16 respondents chose “Elementary school or earlier.”
The total number of respondents for this particular survey question was 88 rather than 244. This is because only those who lost interest in STEM completed this question.
Binary logistic regression
We conducted binary logistic regression analyses to examine the association between participants’ demographic variables, high school experience, timing of and reasons for STEM pathways, and the odds of earning STEM or non-STEM college degrees. Logistic regression has been shown to produce fairly accurate results (Fan & Wang, 1999), and it is considered the best tool for analyzing dichotomous or categorical outcome variables (Osborne, 2015). As the participants of the current study were graduates of hundreds of high schools in regions spreading across multiple Midwestern states, it was unlikely that the data were clustered within certain groups (e.g., schools or districts). This suggests that using multilevel models (e.g., hierarchical linear modeling) for data analysis is unnecessary because the variance in participants’ responses is most likely to be accounted for by individual differences rather than groups. Therefore, we conducted binary logistic regression analyses at the level of individual participants. We conducted the analyses with the Data Analysis and Statistical Software (Stata Version 13).
Missing data and multiple imputation
We conducted multiple imputation to treat the missing data issue. There was a moderate degree of missingness in the data. As shown in Table 2, 19 (76%) of the 24 key variables had various degrees of missing values. The variable Career satisfaction had the highest degree of missing values (i.e., 67/244 = 27%), and the variable Parent education had the lowest rate of missing information (i.e., 2/244 = 1%). We examined the missing data mechanism using a method described by the UCLA Institute for Digital Research and Education’s (2015) Statistical Computing Seminars: Multiple Imputations in Stata, Part 1. With this method, we computed a series of correlations between missingness on a given variable and all other variables in the dataset. The correlations were low in general. This suggests that the data were missing at random (MAR), and therefore met the assumption of multiple imputation (McKnight, McKnight, Sidani, & Figueredo, 2007).
We conducted 40 imputations because several rounds of sensitivity analyses showed that this number of imputations was needed to generate the best estimates of the parameters. In line with the concept of multiple imputation, the analysis results were pooled from results of each of the 40 imputed datasets. After multiple imputation, the number of participants for every variable reached 244. All outcomes reported in the “Results” section were computed from multiply imputed data, with the exception of descriptive information.
Logistic regression models
The main analyses involved three logistic regression models (i.e., Models 1, 2, and 3) in which a set of demographic variables, high school experience variables, and timing of and reasons for STEM pathway variables were entered into the models in three successive stages (see Table 4). STEM degree (i.e., whether a STEM-related college degree or a non-STEM-related college degree was obtained) was the common outcome variable in all three models. Model 1 included six of the nine demographic variables as predictors but excluded the variables Parent education, Born in the United States, and Black. This was because the variable Parent education had a very similar but slighter lower odds ratio (OR) than the variable Parent job as shown in preliminary analyses. We, therefore, excluded Parent education and kept the variable Parent job to minimize the number of predictors in the model. For a similar reason, we excluded the variable Born in the United States and kept the variable Language. The variable Black was excluded for two reasons: (a) none of the five Black students earned a STEM degree (whereas three of the five Hispanic students earned STEM degrees), which led to an absence of standard errors in the preliminary analyses; and (b) we intended to use the variable Hispanic to represent the variable Black to limit the total number of variables in the final analyses. The eight high school experience variables were entered in Model 2. Finally, the four variables of timing of and reasons for STEM pathways were added in Model 3.
Odds of Earning STEM College Degrees: Models 1, 2, and 3.
Note. M1 = Model 1, M2 = Model 2, M3 = Model 3. STEM = science, technology, engineering, and mathematics; CI = confidence interval.
For multiply imputed data, there were no ways to assess the fit of a logistic regression model, such as statistical tests of individual predictors and goodness-of-fit statistics. As a compromise, we evaluated the fit of the three models with the original data. In alignment with Osborne’s (2015) suggestion, we assessed the overall model fit and validated predicted probabilities. Both Model 2 and the final models showed a good model fit.
We excluded the variable Hispanic from Model 2 and the final model because there were only five Hispanic participants, and this variable showed nonsignificant impact on the odds of earning STEM degrees in Model 1.
p < .01. ***p < .001.
Results’ interpretation
In this study, all predictor variables were binary (i.e., having two levels or subcategories), as opposed to multinomial, polychotomous, or polytomous (i.e., having more than two levels). It is relevant to note that this was also the case for the four ethnicity variables: White, Asian, Hispanic, and Black. We created these four binary predictor variables rather than creating a multinomial race variable with four levels. Thus, each of these variables has two levels. For example, the variable White has two dichotomous levels: White (1) and non-White (0). When interpreting the outcomes of logistic regression analyses, the non-White group was considered to be the reference group (also called base group), compared with the White group. Thus, we interpret the results of logistic regression analyses as the odds of earning STEM degrees for White compared with non-White, which could be any of non-White ethnic groups such as Asian, Hispanic, or Black. For example, as shown in Table 4, the odds of earning a STEM degree for White participants was 8.11 times higher than that for non-White participants, holding other variables in the model constant.
Check on collinearity and interactions
Collinearity and interaction effects might exist due to the high correlation between predictor variables (e.g., r ≥ .80) in logistic regression models. This can interfere with the interpretation of analysis results (Pedhazur, 1997). In the current study, we computed the intercorrelations of all 18 predictors in Model 3. The highest correlation was found between the variables White and Asian (r = −.70). The second highest correlation was found between the variables Public school and Independent school (r = −.62). The vast majority of the correlations were lower than .20. Therefore, it was unlikely that the issue of collinearity existed. To further confirm this, we conducted preliminary analyses in which we included interaction terms of all involved predictors. Results revealed no significant interaction terms. We, therefore, chose not to include interaction terms in Models 1 to 3.
Results
Descriptive Results
Demographic variables
Table 2 presents the descriptive information of 24 key variables coded from the original data prior to multiple imputation. Of the 244 participants, 47% reported that they earned a STEM-related college degree, and 53% earned non-STEM-related degrees. Among the 223 who revealed their gender identity, 55% were females, and 45% were males; 60% of the participants identified themselves as White, 29% Asian, 2% Black, and 2% Hispanic; 81% reported that English was their primary language spoken at home, while 19% reported that English was not their primary language spoken at home; 85% of participants were born in the United States and 15% were born outside of the United States; at least one of the parents of 52% of the participants had a doctoral-level education, whereas 48% of participants had parents with education below the doctoral level; and the parents of 70% of the participants had a STEM-related job, whereas 30% reported that their parents had jobs unrelated to STEM.
High school experience and STEM pathways
Descriptive analyses revealed informative findings on participants’ high school experience, their interests in STEM, and what influenced their career choice during high school, which was self-reported by the participants 4 to 6 years after their high school graduation. Table 3 presents descriptive information on six selected survey questions and their responses. Together, these data revealed six findings that might interest families and educational practitioners.
First, most students chose their college major or concentration during high school. The data showed that 50% of the participants reported that high school provided the experiences that were most important in determining their choice of major or concentration, whereas 43% of them reported that college provided the experiences that were most important for them to decide their majors. Less than 7% of participants chose their majors based on their middle school or earlier experiences.
Second, family and friends played a major role in helping high school students make their career choices. Data showed that 41% of the participants reported that their family and friends were responsible for helping them with their career choices during high school, not surprising given the high percentage of students with parents in STEM fields; another 15% turned to mentors within fields they were interested in; 19% of students reported that no one helped them with their career choices, and they did their own career exploration; and 17% of the participants reported that they did not formulate a career goal in high school. Among various career support resources available within and outside of the school, only guidance counselors were found to be helpful, as reported by 17% of the respondents.
Third, when asked about the high school experiences that were most useful to them, the students cited science learning experiences that made them feel comfortable with science concepts (56%), high academic expectations (42%), and acquiring content and background knowledge (38%). Very few students believed that peer-group mentoring (4%) or support services (1%) were the most useful to them.
Fourth, most students perceived acceptance to a college of their choice as their most important academic success in high school. Sixty-two percent of the respondents thought that acceptance to a college of their choice was the most important measure of high school academic success. Other important indicators of academic success included achieving personal goals and/or satisfaction (56%), acceptance to a prestigious college (54%), making family proud (52%), and good SAT or ACT scores (49%). A small portion of respondents believed that publishing research (2%) or preparation to solve local or global problems (12%) was their most important academic success in high school.
Fifth, a good match between students’ natural talents and STEM subjects makes a difference. When asked to indicate possible sources of their initial interest in STEM, more than half of the students (57%) noted that these subjects came easily to them. Other influential sources of interests included enjoying, exploring, and thinking about STEM ideas (54%), encouragement by parents or family members (46%), good grades or awards in STEM (45%), and career aspirations (39%). These data suggest that initial interests, supported and encouraged by family members and rewarded by academic success in classes, are key to cultivating a long-term commitment to STEM study. A very small percentage of students (6%) cited scientific discoveries or famous events as the sources of their initial interest in STEM.
Last, the early college years appeared to be a time when students were most likely to lose their interest in STEM. When asked whether they ever lost interest in STEM and when the loss of interest occurred, 57% cited their freshman or sophomore year of college, 25% lost their interest in high school, 16% lost interest during their junior or senior year of college, and 2% did so between undergraduate graduation and graduate school. These data implicate the college years as a vulnerable time in the STEM pathways, even for academically talented and successful students.
Binary Logistic Regression Analyses
Outcome statistics
Common outcome statistics in logistic regression analyses include correlation coefficients (β), ORs, and predicted probabilities (Huang & Moon, 2013). Because ORs are relatively easier to understand than correlation coefficients, we chose to report ORs as the main outcomes. Other related statistics include standard errors (SE), t values, p values (i.e., p > | t |), and 95% confidence intervals (CI). We conducted statistical tests to examine the differences in the correlation coefficients of different predictors (e.g., White vs. Asian), but we did not report the specific values of correlation coefficients. Variables with higher correlation coefficients have a higher impact on the outcome variable in a logistic regression model.
Furthermore, we computed predicted probabilities (also called predicted margins) for significant predictors in Model 3. Predicted probabilities were calculated with margins command after logistic regression. A predicted probability denoted the average probability of the outcome for one level of a predictor variable while holding other covariates in the model constant. For example, based on the logistic regression in Model 3, the predicted probability was 60% for White and 27% for non-White. It is interpreted that the average probability of earning a STEM college degree was 60% if every subject in the dataset were White, and the average probability of earning a STEM degree was 27% if every subject in the dataset were non-White, while holding other covariates in the model constant. We also reported the results of statistical contrasts for the predicted probabilities of the two subcategories for each significant predictor.
Demographic variables
Model 1 analyzed the association between six demographic variables and the odds of earning STEM college degrees (see Table 4). Results showed that being White or Asian appeared to be a significant predictor of earning STEM college degrees. Specifically, the odds of earning a STEM degree for White students was 6.92 times higher than that for non-White students, holding other demographic variables in the model constant. The difference was statistically significant, t = 3.47, p = .001, 95% CI = [2.32, 20.67]. Similarly, the odds of earning a STEM degree for Asian students was 12.15 times greater than that for non-Asian students, holding other demographic variables in the model constant. Again, the difference was statistically significant, t = 4.27, p < .001, 95% CI = [3.86, 38.27]. Statistical tests of correlation coefficients revealed no statistical difference between those of any two demographic predictors in the model.
High school experience
With the above six demographic variables remaining in the model, eight high school experience variables were entered in Model 2 (see Table 4). As described in the “Method” section, the eight variables depicted three aspects of high school experience: school environment (i.e., whether they attended a public or independent/parochial high school), STEM-related learning (i.e., whether they had hands-on, research, or a field trip experience often), and perceptions of their high school experience (i.e., Prepared, Valuable, and Belonging).
Results showed that the variables White and Asian continued to be a significant predictor of earning STEM college degrees. The odds of earning a STEM degree for White participants was 8.46 times higher than that for non-White participants, holding other predictor variables in the model constant. The difference was statistically significant, t = 3.71, p < .001, 95% CI = [2.74, 26.15]. The odds of earning a STEM degree for Asian students was 14.41 times greater than that for non-Asian students, holding other predictor variables in the model constant. Again, the difference was statistically significant, t = 4.44, p < .001, 95% CI = [4.44, 46.79]. None of the high school experience variables appeared to be associated with the odds of earning STEM college degrees. Statistical tests of coefficients revealed no significant difference between those of any two predictors in the model.
Timing of and reasons for pursuing STEM pathways
With both demographic and high school experience variables remaining in the model, four variables relevant to timing of and reasons for STEM were entered in the final model (i.e., Model 3; see Table 4). Results showed that once again the variables White and Asian continued to be a significant predictor of earning STEM college degrees. The odds of earning a STEM degree for White participants was 8.11 times higher than that for non-White participants, holding other variables in the model constant. The difference was statistically significant, t = 3.26, p = .001, 95% CI = [2.30, 28.57]. The odds of earning a STEM degree for Asian students was 18.91 times greater than that for non-Asian students, holding all other variables in the model constant. The difference was statistically significant, t = 4.36, p < .001, 95% CI = [5.04, 70.96].
The variable regarding interest in STEM emerged as a significant predictor of earning STEM degrees. The odds of earning a STEM degree for those whose interest in STEM sustained or grew over the years was 5.41 times greater than that for those whose interests in STEM permanently or temporarily lost at some point during their time as students, taking into account all other variables in the model. The difference was statistically significant, t = 4.90, p < .001, 95% CI = [2.75, 10.62]. In addition, the variable Select major appeared to be a noteworthy but not a significant predictor. The odds of earning STEM degrees for those who selected college majors early (i.e., high school or earlier) was 2.13 times higher than that for those who selected college majors late (i.e., during college). However, the difference was close but not statistically significant, t = 1.93, p = .054, 95% CI = [0.99, 4.58].
Predicted probabilities
Table 5 presents the predicted probabilities for three significant predictors in Model 3 (i.e., White, Asian, and STEM interests). Results showed that all six predicted probabilities, ranging from 27% to 78%, were statistically different from 0, p < .001. The highest predicted probability (78%) was associated with being Asian. That is, the predicted probability of earning STEM degrees for Asian participants was 78%, compared with that of 33% for their non-Asian counterparts.
Predicted Probabilities of Significant Predictors in Model 3.
Note. CI = confidence interval; STEM = science, technology, engineering, and mathematics.
p < .001.
Statistical contrasts showed the differences between the respective predicted probabilities of the two subcategories of the variables White (i.e., Yes or No), Asian (i.e., Yes or No), and STEM interests (i.e., interests sustained or grew and interests lost) were all significantly different from 0 (see Table 6). Specifically, the predicted probability of earning STEM degrees for White participants was 32% higher than that for their non-White peers. The predicted probability of earning STEM degrees for Asian participants was 46% higher than that for their non-Asian peers. The predicted probability of earning STEM degrees was 33% higher for those whose interests in STEM sustained or grew over time than those who lost their interests in STEM at some point during their time as students.
Contrasts of Predicted Probabilities of Significant Predictors.
Note. CI = confidence interval; STEM = science, technology, engineering, and mathematics.
The contrast/difference between two predicted probabilities was statistically significant, as indicated by the associated 95% confidence interval that did not include 0. This was the same case with the other two variables in the table.
College and beyond variables
To examine the relationship between STEM degrees and several variables during and after college, we conducted separate analyses with three new logistic regression models (i.e., Models 4, 5, and 6). These three new models included White, Asian, Gender, Parent job, and STEM interest as covariates. It is important to note that not all of these three new models had STEM degree as the outcome variable. As Table 7 shows, Model 4 examined the odds of earning a STEM degree for the variable Change major (i.e., whether the student changed majors in college). In Models 5 and 6, however, a STEM degree became a predictor variable, whereas the outcome variable became whether students pursued advanced study at the time of survey, and whether they were satisfied with the direction that they had pursued in their professional life, respectively. None of the variables were significant predictors in the analyses. However, in Model 6, Gender was found to be a significant predictor of Career satisfaction. The odds of feeling satisfied with their professional life for male participants was 0.41 times less likely than that for female participants, t = −1.99, p = .047. 95% CI = [0.17, 0.99]. The predicted probability of feeling satisfied with their professional life for females (91%) was 11% higher than that for males (80%). The predicted probability was statistically different from 0, SE = 0.05, 95% CI = [−0.20, −0.002].
STEM Degrees and Life in College and Beyond.
Note. White, Asian, Gender, Parent job, and STEM interest were used as covariates in the analyses. Gender was found to be a significant predictor of career satisfaction in Model 6. CI = confidence interval; STEM = science, technology, engineering, and mathematics.
p < .05.
Discussion
The current study found that the odds of earning STEM college degrees for Asian or White students were statistically greater than those for their respective counterparts. This is the most robust finding of the current study, consistent across three different models in which demographic factors, high school experiences, and timing of and reasons for STEM pathways were taken into account, respectively. This finding is highly consistent with what some national data and empirical research have shown. For example, the U.S. News/Raytheon 2015 STEM Index shows that gaps between race and gender in STEM are deeply entrenched, although employment and degrees granted in STEM have improved slightly since 2000 (U.S. News & World Report, 2015). Research also frequently notes the concentration and overrepresentation of Asians and Whites in STEM fields (Min & Jang, 2015; Minor, Desimone, Spencer, & Phillips, 2015; Riegle-Crumb, Moore, & Ramos-Wada, 2011). However, unlike most of the previous STEM studies in which male participants were often dominant over females, the sample of the current study consisted of more female (55%) than male participants (45%). Therefore, the current study expands the generalizability of the finding regarding the concentration and overrepresentation of Asians and Whites in STEM fields by adding a special case in which the finding still stands in a sample with an atypical gender composition.
The current study revealed that close to half (47%) of the 244 participants who graduated from non-STEM-focused high schools but participated in supplemental science and/or mathematics courses through a Talent Search program earned STEM college degrees 4 to 6 years after high school. This rate is almost equivalent to 49.8% of the graduates of specialized STEM high schools in the national STEM high school study (see Almarode et al., 2014). This finding confirms what the national STEM high school study had found—specialized STEM high schools and Talent Search programs served equally well as incubators of STEM-talented students. One may argue that both participation in Talent Search programs and enrolling in selective specialized STEM high schools are highly self-selected experiences. These two groups of students might be similar in many characteristics, especially in their high academic achievement levels and learning abilities as well as support and encouragement from parents and access to exceptional teachers and instructional environments. Therefore, their success in earning STEM college degrees is neither surprising nor noteworthy. We acknowledge that the Talent Search participants in the current study and the graduates of selective specialized STEM high schools in the national STEM high school study were top achievers in their local schools or regions. However, this might not fully explain their degree of success, considering that this percentage is much higher than that found for a group of similar-aged gifted peers whose high school SAT or ACT math scores were in the highest quintile in national databases (see Lowell, Salzman, Bernstein, & Henderson, 2009). In other words, the success of Talent Search participants in earning STEM degrees cannot be fully attributable to their high achievement or high abilities. This finding calls for attention to the unique role of Talent Search and supplemental outside-of-school gifted programs in facilitating gifted students’ success in the STEM pathway.
On the basis of this finding, we further argue that supplemental outside-of-school STEM learning opportunities are needed for gifted students to get a richer and mixed dose of STEM education. The current numbers of STEM-focused schools are inadequate to meet the nation’s needs for creating a STEM-capable citizenry, a STEM-proficient workforce, and an abundant supply of future STEM innovators. The National Consortium for Specialized Secondary Schools of Mathematics, Science and Technology (NCSSSMST) currently has approximately 100 member schools across 32 states (NCSSSMST, 2014). Together, these schools serve only a small portion of students (mostly high school students), ranging from 100 to 3,000. Moreover, not all these schools are fully devoted to STEM education. For example, among the 100 member schools of the NCSSSMST, only 60% were comprehensive full-day (i.e., students take all courses in the school), 18% were residential (i.e., students live on campus), and more than one fifth (23%) were half-day programs (i.e., students take STEM courses in the school, then return to another school for other courses; NCSSSMST, 2014). Hence, the recommendation of PCAST (2012) for the creation of at least 200 new specialized STEM high schools and 800 elementary and middle schools in the next decade is well founded. Before this goal becomes realizable, regular schools will remain the place where most students primarily receive their STEM education. For gifted education specialists, there is a great opportunity and need to provide supplemental, enriched, and/or accelerated STEM learning programs to serve gifted students inside or outside of their regular schools.
The current students’ interests in STEM were essential to their success along the STEM pathways. This study found that sustained personal interest in STEM positively and significantly predicted the probability of students earning STEM degrees, above and beyond demographics and all other high school experience factors. This is consistent with findings of some previous research (e.g., Almarode et al., 2014; Maltese & Tai, 2011; Wang, 2013). It is worth noting, however, that 29.6% (60 of 203 respondents, see Table 2 and Note b) of Talent Search participants in the current study reported that their interest in STEM intensified or grew over time. This proportion was slightly higher than 27.9% of the 984 specialized STEM high school graduates in Almarode et al.’s (2014) study. It is probable that students’ experiences in the Talent Search and supplemental outside-of-school gifted programs share some unique features with those of STEM-focused high schools. These include courses on topics not typically part of the high school STEM curriculum, opportunities to pursue areas of interest in independent projects, instruction from academic scientists, access to college science labs and equipment, and interaction with other STEM-passionate students. However, the finding of comparability in terms of pursuit of STEM college majors and intensification of STEM interest over the high school years for the subjects of this study is still noteworthy, given the difference in STEM “dose” between them and students who attend a STEM-focused high school full- or part-time.
Furthermore, although it is not novel to find that students’ interests in STEM were critical for their success in earning STEM degrees, the current study revealed useful information for further understanding what ignited students’ initial interests in STEM, an issue that has gained great attention recently. In particular, this finding validates what VanMeter-Adams, Frankenfeld, Bases, Espina, and Liotta (2014) have found, that the strongest factors that initially sparked students’ interests in STEM were from outside-of-school settings. VanMeter-Adams et al. analyzed survey responses from 129 high school and undergraduate students who had strong interests in STEM and more than 300 hr of research experience in a university summer camp. This study found that more than 60% of the students reported extracurricular encounters as the most important factors that initially cultivated their interests in STEM, compared with 18% to 19.6% reporting classroom experiences and 13.5% to 19.6% reporting hands-on projects. Influences from relatives or friends and childhood experience or encounter with nature or astronomy, for example, stood out among the five categories of extracurricular encounters, noted by approximately 27% and 24% of the students, respectively.
The survey in the current study included a list of 14 options covering a wide range of possible family, school, and outside-of-school factors when asking students to select possible sources of students’ “INITIAL interest” in STEM. Similar to VanMeter-Adams et al. (2014), the current study also found that a higher percentage of students cited parents and family members rather than school-based factors as important sources in cultivating their initial interest in STEM. As Table 3 shows, 46% (113 of 244) of the students mentioned “Encouragement by a parent, guardian or other family member,” 26% (64 of 244) selected “Encouragement by a mentor or teacher,” and 22% (54 of 244) noted “School lessons in STEM” as a source of their initial interest in STEM. It is important to note that although these two studies revealed a similar finding, the current study extends VanMeter-Adams et al.’s finding to a different group of academically talented students.
Families’ influences are not only associated with students’ initial interest in STEM but also with their career interests and choices in high school. The descriptive data showed that 41% of respondents reported that their family and friends were responsible for helping them with their career choices during high school, a finding that is neither new nor surprising. Many previous studies have reached a similar conclusion (e.g., Dabney et al., 2013; Harackiewicz et al., 2012; Maltese & Tai, 2010). However, this finding makes the case that family and parents play a major role in students’ career choices, even for academically gifted students from advantaged families. More than half of the participants in the study (52%) had parents with a doctoral-level education, and 70% of them had parents in STEM-related careers, a likely significant influence on their exposure to STEM topics and knowledge of STEM career paths. Although not investigated in this study due to a small sample size, this finding suggests the need to understand the nature of STEM pathways for students from less advantaged family backgrounds and students without parents working in a STEM field. For those students, navigating the STEM pipeline can be daunting, and support from educators and mentors is even more critical.
Finally, this study uncovered a pressing issue—the need to augment and improve career guidance for students, even for relatively advantaged gifted students. Results showed that although 41% of the participants received help from their family or friends with their career choices, 19% of them had no help and explored their career choices alone, and 17% did not identify a career goal during high school, which may thwart their ability to enter top-notch STEM programs. Furthermore, only 17% of the participants found guidance counselors helpful among available career support resources within or outside of their school. This suggests that an alarming portion of students lacked support to learn about and explore STEM careers, which is essential for inspiring and enabling more students to pursue STEM pathways. It is, therefore, imperative for educators to devote more effort to implementing effective strategies to guide students’ career exploration and decisions, which can include job shadowing, internships, career days, and counseling. Furthermore, to provide effective support and interventions for gifted individuals, educators ought to give special consideration to issues that might particularly affect gifted students’ career planning and development, such as multipotentiality, perfectionism, underachievement, and pressure toward particular careers from family and others (see Greene, 2006; Maxwell, 2007; Muratori & Smith, 2015).
Future Research
Future research can explore a series of questions regarding STEM interests of gifted populations. First, why were some students’ interests in STEM sustained and even intensified, whereas others lost their interest at some point? Second, in what specific ways did the parents and teachers of students who sustained or intensified their interests influence them in comparison with students who lost their interest? Third, how can high school coursework and/or supplemental enrichment or accelerated courses sustain and grow students’ interests in STEM? What are the features of these experiences that influence student interest and commitment? Fourth, what is the role of peer relationships, both at a regular school and in outside-of-school supplemental courses, in affecting students’ interests in STEM? In sum, as students’ personal interests in STEM significantly predict their success along the STEM pathways, it is critical to understand how and in what ways we can encourage and grow it, both within and outside-of-school environments.
Study Limitations
Findings of the current study must be understood in light of an important limitation. That is, this study relies on data collected via a survey that was primarily designed for the graduates of selective specialized STEM high schools—the vast majority of the participants in the national STEM high school study. The survey focused on participants’ high school experiences, especially STEM learning experiences, such as the nature and amount of laboratory experience and the sources of interest in STEM. This served as the main purpose of the national STEM high school study. However, the survey lacked questions that inquired about experiences or issues unique to the past participants of Talent Search programs, such as the reasons for attending regular rather than specialized STEM high schools, or the frequency or types of supplemental (e.g., enriched or accelerated) courses students took through Talent Search programs. As a result, although the key background of the participants in the current study was their experience in a Talent Search program and their participation in supplemental enriched and accelerated learning activities, we were not able to study the specific aspects of these experiences and participation and their association with STEM outcomes.
Conclusion
This study provides new empirical evidence on the prevalence of Asian and White students in STEM majors in college, and the importance of personal interests and family influences for students’ decision to enter or success along a STEM pathway. It presents an important case that specialized STEM high schools are not the only effective way to develop STEM talents of gifted students. Gifted programming, such as Talent Search, and many other forms of supplemental enriched or accelerated learning opportunities might be a viable and less costly alternative for STEM education pathways. Echoing what was concluded in the national STEM high school study, the current study provides additional evidence supporting the PCAST’s (2012) recommendation for the development of a wide range of high-quality STEM-based additional learning opportunities, including summer and afterschool programs, STEM contests, and other informal activities. The field of gifted education ought to seize these great opportunities in providing supplemental enriched and/or accelerated STEM learning programs to serve gifted students inside or outside of their regular schools.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.