What Boosts Talent Development? Examining Predictors of Academic Growth in Secondary School Among Academically Advanced Youth Across 21 Years

Abstract

We examined 482,418 students who took the ACT in the seventh grade and again in high school, taking an exploratory analytic approach to examine academic growth trends from 1996 to 2017. Predictors included sociodemographics, interests, high school (HS) characteristics, HS coursework and GPA, and extracurriculars, which explained 25% of the variance in academic growth. Overall, growth improved from 2005 to 2017, but growth for low-income and Hispanic students was stagnant. Catholic and private school students had the highest growth; homeschooled and high-poverty public school students had the lowest. High growth was associated with STEM (science, technology, engineering, mathematics) elective HS courses and advanced AP, accelerated, and honors courses. Students with investigative and conventional interests had higher growth. Some extracurriculars had significant relationships with academic growth, though the effects were small.

Keywords

academically gifted academic talent

Talent development is important because it can contribute to individual development, personal fulfillment, later educational and occupational achievement, and societal innovation (Lubinski & Benbow, 2006; Subotnik, Olszewski-Kubilius, & Worrell, 2011). An innovation pipeline for data science and artificial intelligence talent (Executive Office of the President of the United States, 2016; The Networking and Information Technology Research and Development Program, 2016), a desire to increase gross domestic product (Rindermann & Thompson, 2011), and concerns over the development of “human capital” (Heckman, 2000) needed to maintain national competitiveness (Augustine, 2005; Hanushek & Woessmann, 2015; National Science and Technology Council, 2016) are very pragmatic policy concerns to which the talent development process of gifted students connects (Benbow & Stanley, 1996; Wai & Worrell, 2016). Perhaps the most important outcome of adequate talent development is the personal and optimal flourishing of gifted students so that they can realize their full potential in all aspects (Subotnik et al., 2011).

Despite the importance of developing the talent of academically advanced or gifted students, very little is known about what best boosts talent development. This study adds to the literature by using an existing talent search process and following students over time to see what programs and experiences appear to be most associated with academic growth. We define academic growth in this paper to be the relative change in academic achievement, as measured by a standardized test, from Grade 7 to Grade 11/12.

One of the major talent search processes in the United States is the talent search model, through which more than 100,000 seventh grade students each year take a college entrance exam designed for juniors and seniors such as the ACT (ACT, 2017) or SAT (College Board, 2016; e.g., Assouline & Lupkowski-Shoplik, 2012). These students typically qualify initially for participation in a talent search by being identified as academically advanced (i.e., scoring in the top 5%) on a grade-level test. These academically advanced students then take the ACT or SAT in the 7th grade to determine their academic readiness for advanced educational programming matched to their level and pattern of domain-specific strengths. Due to the 7th-grade talent search coupled with the need for students to take a college admissions exam in high school, many students who take the ACT in the 7th grade do so again in 11th or 12th grade (Allen, 2016), providing a unique leverage point to investigate academic development through the secondary school years. The talent search has operated for at least two decades, providing an historical window into possible improvements in academic growth among academically advanced or gifted youth. Understanding what boosts talent development across time via the talent search process is useful because of the potential for differences in cultural context or history to influence the presence and degree of talent development. If changes have or have not occurred across time, it is important to then try to understand why.

ACT and SAT scores are important for access to college (e.g., Dynarski, 2018; Hyman, 2017) and for predicting first-year college outcomes as well as later educational and occupational success (e.g., Kobrin, Patterson, Shaw, Mattern, & Barbuti, 2008; Kuncel, Hezlett, & Ones, 2004; Schmitt et al., 2009). A better understanding of which factors are associated with score growth has the potential to lead to interventions that can improve the achievement of academically advanced students and potentially all students. Prior research using the ACT as a measure of academic growth revealed that additional high school coursework and improved course performance are associated with enhanced academic growth for the general population of students (Sawyer, 2008) and that participation in a summer academic program enhanced academic growth for academically advanced students (Schiel, 1998). Prior research has focused primarily on high school coursework and grades, so the potential influence of sociodemographic variables, interests, high school characteristics, and extracurricular activities remains relatively unexplored territory. Advanced placement (AP) coursework in particular is of importance for the gifted, given its highly visible role in being a form of academic enrichment (Assouline, Colangelo, VanTassel-Baska, & Lupkowski-Shoplik, 2015) and its importance in talent development (Bleske-Rechek, Lubinski, & Benbow, 2004); thus, AP should be more thoroughly studied in the context of an array of other potentially important variables for talent development. Research on extracurricular activities among the gifted is important because how gifted students allocate their time across various activities, both inside and outside of school and across academic and nonacademic activities, affects time spent learning (e.g., Makel, Wai, Putallaz, & Malone, 2015).

In particular, there is something to be learned by leveraging and more deeply exploring the outcomes of established and existing processes—such as the seventh grade talent search, a fixture in the education of many gifted students in the United States—to learn more about whether and how gifted students develop over time, and perhaps to better understand what the intended or unintended consequences of such processes are (Assouline & Lupkowski-Shoplik, 2012). Access to college remains an important hurdle for gifted students from African American, Latino, rural, and low-income populations and this access may in part be influenced by access to the seventh grade talent search process, in addition to the many other factors that may influence academic achievement or other aspects of growth before and between the 7th and 11th/12th grades (Olszewski-Kubilius, 1998).

One of the many important contributors to academic growth differences between students from different demographic subgroups may be the type of school environment to which they have access (e.g., Rudner, 1999; Wenglinsky, 2007; West & Woessman, 2010). A longstanding debate has surrounded the relative costs and benefits of different environments in K-12 schooling; thus, it is important to consider the types of school environments typically studied in other contexts and the roles these may play in the growth of disadvantaged students. Investigating differences in academic growth for different subgroups of interest, such as disadvantaged and underrepresented populations, is important to understand how such populations have fared over time broadly but also how such populations have fared relative to other subgroups. If things have changed or have not changed across time, it is important to then try to understand why. Better understanding the academic growth of the disadvantaged subpopulations who have had access to the talent search is an important stepping-stone to potentially helping disadvantaged students more broadly (Wyner, Bridgeland, & DiIulio, 2007).

The study of interests has a long and established history in industrial/organizational psychology (Su & Rounds, 2015). Interests have been shown to be useful predictors of educational and occupational patterns in longitudinal studies of gifted populations (Lubinski & Benbow, 2006). Additionally, extracurricular activities are important to consider (Cooper, Valentine, Nye, & Lindsay, 1999) given that students, including gifted students—perhaps due to personal interests, available resources, or other factors—allocate their time differently across different types of activities, whether inside or outside of school (Makel et al., 2015). There are also numerous factors that influence learning, such as abilities, interests, motivation, persistence, and quality of education (e.g., see Benjamin, 2018, for a review), which suggests that including as many variables as possible in an investigation of academic growth would be worthwhile.

The period between 7th grade and 11th/12th grade includes many opportunities for academic development, both inside and outside of school. Therefore, it is important to examine many variables that might contribute to academic growth, with some of these factors being amenable to intervention. We include numerous variables that are relevant to talent development, including many that have been traditionally studied regarding talent development in the literature. However, more broadly we allow multiple variables to compete empirically to give us hints at which variables may be more or less important for talent development, rather than entering with strong hypotheses. In other words, in this study we leveraged a unique sample and took an exploratory analytic approach: Using data from 22 academic years, we follow students from 7th through 11th/12th grade, looking at what factors best predict growth in ACT scores across this time span. We examine potential predictors that might contribute to academic growth, including sociodemographic variables, interests, high school characteristics, high school coursework and GPA, and extracurricular activities. We attempt to explain two types of changes: students’ ACT score change from 7th grade to 11th/12th grade (academic growth), and group improvement in academic growth from 1996 to 2017 (growth trend).

Research Questions

We address the following research questions:

Research Question 1: Has growth in academic achievement from 7th to 11th/12th grade improved among academically advanced youth in the past two decades?

Research Question 2:. Has growth in academic achievement from 7th to 11th/12th grade improved among academically advanced youth within special subgroups of interest?

Research Question 3: Is variation in academic growth among academically advanced youth explained by sociodemographics, high school characteristics, coursework taken, high school GPA, Holland-type vocational interests, or extracurriculars?

Research Question 4: To what extent have predictors of academic growth among academically advanced youth changed over the past two decades?

Deidentified data obtained from ACT databases of testing records are the primary source of data for this study. As described next, students’ ACT test scores from 7th grade and 11th or 12th grade were used to measure academic growth. By analyzing these data across several cohorts of students, we are able to examine trends in academic growth. As part of the ACT registration process, students provide demographic information and high school course grades; the information obtained from the later ACT test administration (Grade 11 or 12) provides the basis for our analysis of predictors of academic growth.

Method

Sample

The sample consists of 482,418 students who took the ACT test in 7th grade and again in 11th or 12th grade and were projected to complete high school between 1996 and 2017. A small number of students who took the ACT test with special accommodations in 7th grade but not 11th/12th grade (or vice versa) were excluded from the analysis (n = 3,339, 0.69%). The vast majority of students in the sample (96%) sent their 7th grade ACT score results to a major talent search program. The mean 7th grade ACT Composite score was very similar for students who sent their scores to a talent search program (17.7) and those who did not (17.8). Thus, we included all students in the sample and considered them academically advanced, even if they did not send their scores to a talent search program.

The sample can be compared with the general population of 11th grade students (as of 2016) in the United States on gender, race/ethnicity, geographic region, school type, and school locale (Table 1). The sample and population are both evenly split on gender. Relative to the population, the sample has a higher percentage of White students (81% for sample, 52% for population), and lower percentages of African American (4% vs. 15%), Asian (3% vs. 6%), and Hispanic students (4% vs. 24%). The sample is mostly from the Midwest (36% for sample, 22% for population) and South (62% for sample, 38% for population), with very little representation from the Northeast (1% for sample, 17% for population) and West United States (2% for sample, 23% for population). Students from public schools with a higher poverty concentration (larger percentage of students eligible for free or reduced lunch) are not well represented by the sample. A small percentage of students in the sample (0.3%) were home-schooled when they tested in high school. Relative to the population, the sample has a larger share of students attending schools in rural (22% vs. 13%) and town (17% vs. 11%) locales and a smaller share of students attending schools in suburban (32% vs. 40%) and city (29% vs. 32%) locales.

Table 1.

Comparing the Sample to the Population of U.S. Students in 11th Grade and ACT-Tested Population.

Variable	Sample cohorts (%)			11th grade population (%)	ACT-tested population (%)
Variable	Total 1996-2017	Early 1996-2000	Recent 2013-2017	11th grade population (%)	ACT-tested population (%)
Gender
Female	50.3	52.1	49.2	49.3	51.6
Male	49.7	47.9	50.8	50.7	46.3
Unknown					2.1
Race/ethnicity
African American	4.1	2.7	5.4	14.9	12.7
Asian	2.8	1.6	4.6	5.5	4.7
Hispanic/Latino	3.8	1.5	7.3	24.3	17.1
White	80.7	81.8	74.1	51.6	52.3
Other	2.5	1.4	3.9	3.7	5.4
Unknown	6.0	11.1	4.7		7.8
Region^a
Midwest	35.6	41.9	27.8	22.2	27.6
Northeast	0.6	0.1	1.3	16.6	8.5
South	61.7	55.3	68.5	37.8	44.4
West	2.1	2.7	2.4	23.3	19.6
School type
Catholic	9.3	7.4	9.5	3.1	4.3
Home	0.3	0.1	0.6	NA	0.9
Private	8.8	6.5	10.7	4.2	5.7
Public, <20% FRL	28.3	39.6	18.9	14.8	17.1
Public, 20%-40% FRL	28.4	28.8	27.6	22.9	23.8
Public, 40%-60% FRL	17.3	11.8	22.0	22.9	22.5
Public, 60%-80% FRL	5.3	3.8	7.5	15.3	12.6
Public, >80% FRL	1.1	1.1	1.4	9.5	8.4
Public, FRL unknown	0.0	0.0	0.0	3.7	3.0
Other	1.1	1.0	1.8	3.6	1.6
School locale
Rural	21.5	30.5	16.2	13.3	14.7
Town	17.0	20.5	13.9	10.7	12.0
Suburban	32.3	24.5	37.3	39.8	38.1
City	29.2	24.6	32.6	32.4	28.9
Unknown				3.8	6.3

Note. Population percentages for gender and race/ethnicity reflect public school students only; FRL = free or reduced-price lunch; NA = not applicable.

Definition of geographic region used by the U.S. Census Bureau.

The sample can also be compared with a recent ACT-tested population (all high school graduates of 2017 who took the ACT test). Relative to the ACT-tested population, the sample has more males and more white students and fewer African American, Asian, and Hispanic students (Table 1). The geographic disparity between the sample and ACT-tested population is similar to, but not as extensive as, the disparity between the sample and general population. One reason for the sample’s sparseness in the Northeast and West is that the ACT test was less prominent in those regions during the study period (see “Where the SAT and ACT dominate”; Lewin, 2013). Another reason is that 7th-grade talent searches are more likely to utilize the ACT test in the Midwest and South and more likely to utilize the SAT test in the West and Northeast.

Table 1 shows the makeup of the sample over the entire study period and also for the earliest cohorts (1996-2000) and the most recent cohorts (2013-2017). This lets us examine how the sample composition has changed over the study period. For example, 50.3% of the sample is male, and the percentage of males has trended upward (47.9% for 1996-2000, 50.8% for 2013-2017). The sample has trended toward more racial/ethnic minority students (e.g., +5.8% Hispanic) and fewer White students (−7.7%). The sample’s representation from the South increased (+13.2%), with a decrease in the Midwest region (−14.1%). Students in the sample increasingly attended Catholic (+2.1%) or private schools (+4.2%), were home-schooled (+0.5%), or attended public schools with higher poverty levels; fewer students in the sample attended low-poverty public schools (−20.7%). Over time, more students in the sample attended schools in suburban (+12.8%) and city (+8.0%) locales, and fewer attended schools in rural (−14.3%) and town (−6.6%) locales. The changes in the sample over time reflect trends in the nation’s demographics, changes in the general ACT-tested population, and possible changes over time in who participates in the 7th-grade talent search by taking the ACT test.

Measure of Academic Growth

Academic growth is measured by students’ performance on the ACT tests they took in 7th grade and again in high school (11th or 12th grade). The tests are designed to measure the academic skills necessary for education and work after high school and cover the general content areas of college and high school instructional programs (ACT, 2017). The multiple-choice portion of the test (215 questions) determines subject area scores in English, mathematics, reading, and science. The ACT Composite score is the average of the four ACT subject area scores from the multiple-choice portion of the test and is the focus of this study. Academically advanced 7th graders, who presumably have not taken specific college-prep courses, score about 0.5 standard deviations below ACT-tested high school graduates, on average (ACT, 2016; Allen, 2016). Research has shown that 7th-grade scores on a college admissions test (the SAT) predict later long-term educational and occupational achievement (e.g., Park, Lubinski, & Benbow, 2007; Wai, Lubinski, & Benbow, 2005).

ACT scale scores range from 1 to 36 and have reliabilities ranging from .85 (for the science test) to .97 (for the Composite; ACT, 2017). Several new forms of the ACT tests are developed each year and are constructed to adhere to the same content and statistical specifications. Scale scores are equated across test forms using randomly equivalent groups and spiraling of multiple test forms (including an anchor form). The equating process ensures that scale scores are comparable across test forms and test dates (ACT, 2017). For this study, we used students’ 7th-grade scores and last scores from high school. The correlation of 7th grade and later ACT Composite scores was r = .74, suggesting that using the ACT for a younger but academically advanced population results in scores that agree well with those from the typical use of ACT scores for students in Grade 11 or 12. Additional validity evidence for using ACT scores for advanced 7th grade students is documented elsewhere (ACT, 2017). The mean (standard deviation [SD]) number of months between the 7th-grade test and high school test was 55.1 (3.5), or about 4.6 years. Although some students take the ACT test multiple times in high school, we chose to only use students’ last scores to maximize the length of the follow-up period and to better capture effects of intervening variables (e.g., coursework, grades, and extracurricular activities) that could manifest throughout high school.

Because ACT scale scores are equated, a student’s score from 7th grade can be directly compared with their score from 11th or 12th grade. One measure of growth is the gain score (later ACT score – 7th grade ACT score). This measure is attractive for its simplicity and intuitive appeal but suffers from ceiling effects (students with high 7th grade scores have less room to grow), regression to the mean (e.g., the lowest scoring students in 7th grade have the largest mean gains), and high standard error of measurement.

The residual gain score (Castellano & Ho, 2013) is an alternative to the simple gain score and is calculated as the residual from a regression of the later test score on the earlier test score. The residual gain model belongs to the same family of conditional status models as other popular growth models, including the student growth percentile model (Betebenner, 2009) and value-added models (Castellano & Ho, 2015). The regression model can be a simple linear model, a higher order polynomial model, or even a categorical model. Because of our large sample (and sufficient sample size at each 7th grade score point), we used a categorical model. Relative to gain scores, residual gain scores often have lower standard error of measurement. Demonstrating this model, Figure 1 plots the mean 11th/12th grade ACT Composite score for each possible 7th grade ACT Composite score (solid black line). For example, among students who had an ACT Composite score of 15 in 7th grade, the mean ACT Composite score in grade 11/12 was approximately 24. Therefore, a student who scores 15 in 7th grade is expected to score a 24 in 11th/12th grade. A student who scores 15 in 7th grade and 30 in 11th grade (depicted by the star in Figure 1) has a gain score of 15 (30-15) and a residual gain score of 6 (30-24).

Figure 1.

Mean 11th/12th grade-ACT Composite score by 7th-grade ACT Composite score.

The residual gain model addresses regression to the mean because average gain is allowed to vary across the score scale. In Figure 1, mean gain scores are represented by the distance between the solid black line and the line of no gain (the dashed black line); regression to the mean is illustrated by the distance between the two lines shrinking as grade 7 ACT Composite score increases. The ceiling effect is represented by the “room to grow” distance (e.g., the difference between the top of the grade 11/12 score scale and the dashed black line) shrinking as Grade 7 ACT Composite score increases. Academically advanced students are more likely than general population students to hit the ACT score ceiling in Grade 11/12, but it is still a rare occurrence: In our sample, 1,181 students (0.24%) had an ACT Composite score of 36.

The residual gain model leads to a normative interpretation of growth. For example, a residual gain score of 0 is obtained when a student’s later test score is equal to the mean among all students who had the same seventh grade test score. A positive residual suggests greater than expected growth, and a negative residual suggests less than expected growth. Growth is calculated with respect to the “norm” in the sample of 482,418 academically advanced youth. Because the same norm is applied to all students in the sample, growth trends over time and predictors of growth can be examined.

In addition to seventh grade test score, the residual gain model can accommodate other covariates in the regression equation. We used the number of months between the two tests as a covariate because students are expected to learn more with longer intervening periods. The inclusion of additional covariates does not change the interpretation of the residual gain score as a normative measure of growth but adjusts the measure for potential confounding variables. Across all students and cohorts, the mean and standard deviations were 17.7 (3.1) for the 7th-grade ACT Composite score, 26.7 (4.1) for the 11th/12th-grade ACT Composite score, 9.0 (2.7) for the gain score, and 0.0 (2.7) for the residual gain score. Whereas “conditional status” is a more precise descriptor of what is measured by the residual gain score, we use the term “growth” throughout the paper for simplicity. The correlation of simple gain score (change in absolute performance) and residual gain score (change in relative performance) was r = .99, suggesting that there is little practical difference between these two types of growth measures.

Predictors

Sociodemographic Variables

Gender, student-reported family income level, parent education level, and race/ethnicity were collected when students registered for the ACT in high school and were used as predictors and to identify students in special subgroups of interest. Student-reported family income level was collected as an ordinal variable. To account for wage inflation during the study period, income was categorized relative to the median household income in the United States during the student’s cohort year: lower income (<75% of median), middle income (75%-125% of median income), and higher income (>125% of median). For students in the 2011-2017 cohorts, parental education level was collected, and we used categories of high school or less, some college but less than a bachelor’s degree, and bachelor’s degree or higher. Race/ethnicity was categorized as African American, Asian, Hispanic, White, and other (including Native American and two or more races). The ACT registration form also enables researchers to collect data on disability status (students are asked whether they have a disability that requires special provisions from the educational institution) and whether English is the primary language spoken in the home. These data were used to identify additional special subgroups of interest.

High School Characteristics

Data on high school characteristics were obtained from the National Center for Education Statistics Common Core of Data (Glander, 2016) and the Market Data Retrieval school database (http://schooldata.com/). Variables included school category (Catholic, private, public, or other), percentage of students eligible for free or reduced-price lunch (FRL, available only for public schools), class size, and locale (rural, town, suburban, or city). A school category variable was created that incorporates school type and FRL percentage (for public schools, using 5 intervals of equal width for FRL), resulting in nine categories: Catholic, private, home, other (e.g., state or county-operated schools), public < 20% FRL, public 20% to 40% FRL, public 40% to 60% FRL, public 60% to 80% FRL, and public > 80% FRL.

High School Coursework and GPA

Self-reported high school coursework and grades were collected when students registered for the ACT test in high school. For 30 different courses, students were asked if they (1) have taken the course (or are currently taking the course), (2) have not taken the course but plan to later, or (3) have not taken the course and will not take it later. For this study, students were classified as having taken a course if they marked option 1. Some high school courses (e.g., English 9, Algebra I, biology, and U.S. history) are taken by virtually all students because of prerequisites or high school graduation requirements and so are of less interest as predictors of academic growth. We examined 18 elective courses, coded as binary indicators, as predictors of academic growth (see Table 4 for list of courses).

When students registered for the ACT test, they were also asked whether they have taken advanced placement, accelerated, or honors courses in English, mathematics, social studies, natural sciences, or foreign languages. Binary indicators for each type of advanced coursework were used as predictors of academic growth.

High school GPA was determined by averaging grades reported by students across up to 23 core high school courses. Although the high school GPA measure is based on student self-report, it is highly correlated with high school GPA obtained from high school transcripts (r = .84; Sanchez & Buddin, 2016) and is used as a predictor of academic growth.

Vocational Interests

Holland’s (1997) theory of vocational choices proposed there are six work environments that correspond to six personality types: Realistic, Investigative, Artistic, Social, Enterprising, and Conventional. The ACT Interest Inventory (ACT, 2009) measures the six dimensions corresponding to Holland personality types. Students can complete the Interest Inventory when they register for the ACT test. Each item describes an activity (e.g., “Explore a science museum”), and students are asked to indicate if they like, dislike, or are indifferent to doing the activity. Two versions of the Interest Inventory, known as UNIACT-R and UNIACT-S were used during the study period. UNIACT-S was introduced in 2004 and includes 72 items that are a subset of the 90-item UNIACT-R (introduced in 1989). Table 2 provides the names and descriptions of the six scales, and the items are available online (ACT, 2009). Internal consistency reliability estimates for the Interest Inventory scores range from 0.87 to 0.92 for the UNIACT-R and 0.84 to 0.91 for UNIACT-S (ACT, 2009). The six ACT Interest Inventory scores obtained in high school were used as predictors of academic growth. Although interest scores obtained in 7th grade were not used due to high rates of missing data, interests have been shown to be reasonably stable from 7th grade to age 28 for a similar population of academically advanced students (Lubinski, Benbow, & Ryan, 1995).

Table 2.

ACT Interest Inventory Scales.

ACT interest inventory scale (corresponding Holland type)	Description (ACT, 2009, p. 3)
Science and Technology (Investigative)	Investigating and attempting to understand phenomena in the natural sciences through reading, research, and discussion
Arts (Artistic)	Expressing oneself through activities such as painting, designing, singing, dancing, and writing; artistic appreciation of such activities (e.g., listening to music, reading literature)
Social Service (Social)	Helping, enlightening, or serving others through activities such as teaching, counseling, working in service-oriented organizations, and engaging in social/political studies
Administration and Sales (Enterprising)	Persuading, influencing, directing, or motivating others through activities such as sales, supervision, and aspects of business management
Business Operations (Conventional)	Developing and/or maintaining accurate and orderly files, records, accounts, etc.; following systematic procedures for performing business activities
Technical (Realistic)	Working with tools, instruments, and mechanical or electrical equipment. Activities include building, repairing machinery, and raising crops/animals

Extracurricular Activities

Students were also asked which types of extracurricular activities they participated in during high school. We examined 13 types of extracurricular activities, coded as binary indicators, as predictors of academic growth (see Table 4 for the list of activities).

Imputation of Missing Predictor Variables

Student-level predictor data (sociodemographic, high school coursework and GPA, vocational interests, and extracurricular activities) were provided voluntarily by students and can be missing. Missing data rates were 22% for student-reported family income level, 6% for race/ethnicity, ~0% for gender, 9% for high school GPA, 4% to 15% for specific high school courses, 13% for the advanced high school coursework indicators, 14% for vocational interest scores, and 12% for extracurricular activities.¹ Parent education level was only collected for students in cohorts of 2011 and later and was missing for 17% of those cases. To enable regression analyses based on the entire sample, multiple imputation (Rubin, 1996) was used to impute the predictor variables. Instead of filling in a single value for each missing value, multiple imputation replaces each missing value with a set of plausible values that represent the uncertainty about the missing value. We used multiple imputation to produce 50 complete data sets, and the regression models were fit separately for each of the 50 data sets. The SAS PROC MIANALYZE procedure was then used to combine results across the 50 models. This procedure uses the parameter estimates and associated standard errors for each of the 50 regression models and derives valid univariate inference for these parameters.

Other data used to identify special student subgroups (disability status and whether English is the primary language spoken in the home) had much higher missing rates. We chose not to impute these indicators and did not include them as predictors in multiple regression models. However, we still used these indicators to identify students in the respective subgroups (students with disability, students from a non-English speaking home) for analyzing trends in academic growth by subgroup.

Statistical Analysis

We now provide details on linear regression models used to address each research question (RQ). For RQ1 (Has growth in academic achievement from 7th to 11th/12th grade improved among academically advanced youth in the past two decades?), we regressed the academic growth score on cohort year, treating cohort year as a categorical variable. The per-year change in growth was estimated using a linear contrast of the cohort year means. We fit the trend models using data from all cohorts (1996-2017; N = 482,418).

RQ2 (Has growth in academic achievement from 7th to 11th/12th grade improved among academically advanced youth within special subgroups of interest?) used the same approach as used for RQ1, except that the linear regression models were fit separately for each subgroup of interest (female, male, African American, Hispanic, low income, with disability, and from a non-English speaking home).

RQ3 (Is variation in academic growth among academically advanced youth explained by sociodemographics, high school characteristics, coursework taken, GPA, Holland-type vocational interests, or extracurriculars?) is addressed by regressing the academic growth score on the entire set of predictor variables (e.g., residual gain score = sociodemographics + high school characteristics + coursework + GPA + Holland-type vocational interests + extracurriculars). This analysis was conducted using all cohorts (1996-2017; N = 482,418).

Note that RQs 2 and 3 both include comparisons of subgroups. RQ3 examines predictors of growth (including subgroup membership) from grade 7 to grade 11/12 whereas RQ2 compares subgroups’ improvement in growth (across cohorts).

RQ4 (To what extent have predictors of academic growth among academically advanced youth changed over the past two decades?) is addressed by regressing the growth score on the entire set of predictor variables separately for the five earliest cohorts (1996-2000; N = 83,176) and the five most recent cohorts (2013-2017; N = 129,034). The results from the two periods were then contrasted.

RQs 3 and 4 involve interpretation of several regression coefficients. To facilitate these interpretations, the continuous variables (residual gain ACT Composite score, school class size, high school GPA, and Holland-type vocational interests) were standardized to have mean 0 and standard deviation 1. All categorical variables (school type, school locale, race/ethnicity, income level, parent education level, elective and advanced coursework, and extracurricular activities) were dummy-coded.

Results

Has Growth Improved?

To address RQ1, we first plotted the mean growth scores (residual gain ACT Composite scores) by cohort year (Figure 2). By definition, the mean growth score is 0 in the total group. There was a general downward trend in growth from 1996 to 2005, and the mean growth score was at its lowest point (−0.60) for the 2003 cohort. From 2005 to 2017, there was a consistent upward trend in growth, and the mean score was at its highest point (0.53) for the 2016 cohort. Because the trend was different for 1996 to 2004 and 2005 to 2017, we conducted three separate trend tests: 1996 to 2017 (entire study period), 1996 to 2004 (early period of the downward trend), and 2005 to 2017 (later period of the upward trend).

Figure 2.

Mean ACT Composite residual gain scores for total group, by cohort.

Over the entire study period, mean growth from 7th to 11th/12th grade improved by 0.035 units per year for academically advanced youth (Table 3, p < .001). This improvement can be translated to ACT Composite score points over the entire study period by multiplying the coefficient by the number of years (21), translating to 0.74 (0.035 × 21 years) ACT Composite score points from 1996 to 2017. The improvement can also be translated to an effect size measure, such as Cohen’s d. For example, the standard deviation of ACT Composite scores among all ACT-tested high school graduates is 5.6 (ACT, 2016), so the improvement is comparable to an effect size of about 0.13 (0.74/5.6), which is similar to Cohen’s benchmark for a “small” effect (Cohen, 1988). For the later period (2005-2017), mean growth improved by 0.062 units per year and for the earlier period (1996-2004), mean growth decreased by 0.052 units per year. The results suggest that academic growth from 7th to 11th/12th grade among academically advanced youth has improved over the past two decades, with the improvement occurring over the past 13 years. The average gain (difference score) in ACT Composite score from grade 7 to grade 11/12 was 9 points, which translates to about two points per academic year. This provides a benchmark for comparing the improvement with growth that is typically obtained from one year of instruction: The improvement of 0.74 ACT Composite score points over the 22-year period is comparable to students in 2017 having received an extra 0.37 year of instruction, relative to students in 1996 (calculated as 0.74/2.00).

Table 3.

Testing for Trends in Academic Growth Among Academically Advanced Students, by Subgroup.

Subgroup	N	Trend tests
		1996-2017		1996-2004		2005-2017
		β	Standard error	β	Standard error	β	Standard error
Total group	482,418	*0.035	0.001	*−0.052	0.003	*0.062	0.001
Female	242,538	*0.031	0.001	*−0.044	0.003	*0.058	0.002
Male	239,880	*0.036	0.001	*−0.062	0.004	*0.063	0.002
African American	20,895	*0.011	0.004	*−0.064	0.016	*0.055	0.007
Hispanic	20,124	0.007	0.005	*−0.111	0.020	*0.029	0.007
Low income	71,959	*0.005	0.002	*−0.037	0.008	*0.021	0.004
Student with disability	11,489	*0.028	0.005	*−0.072	0.017	*0.062	0.010
Non-English-speaking home	8,260	0.001	0.004	−0.009	0.031	0.023	0.012

Has Growth Improved for Subgroups of Interest?

To address RQ2, we examined separately the growth trends for the following student subgroups: female, male, African American, Hispanic, low-income students, students with disabilities, and students from non-English speaking homes (Figure 3). Although the total sample is predominantly White, subgroup sample sizes are still robust, ranging from N = 8,260 for students from non-English speaking homes to N = 20,895 for African Americans. From 1996 to 2017, significant im-provement in growth was observed for females, males, African Americans, and students with disabilities but not for Hispanic, low-income, and students from non-English speaking homes (Table 4). Relative to the total group (β = 0.035), the improvement in growth was less pronounced for African American students (β = 0.011). The improvement for males (β = 0.036) was similar to the improvement for females (β = 0.031); the improvement for students with disabilities was slightly lower (β = 0.028) than the total group improvement. Each subgroup showed more improvement during the later time period (2005-2017) relative to the earlier time period (1996-2004). The largest differential was observed for Hispanic students and students with disabilities; the smallest differential was observed for students from non-English speaking homes and low-income students.

Figure 3.

Mean ACT Composite residual gain scores for total group and student subgroups, by Cohort.

Predictors of Growth

For RQ3, linear regression was used to relate the full set of predictors to the measure of academic growth using the total sample and all cohorts of students (see Table 4, results for all cohorts). Due to the large sample size, most predictors were statistically significant, even if the regression coefficient was small. We consider coefficients of 0.05 and larger as substantively meaningful and most worthy of discussion. A 0.05 change in the dependent variable (standardized residual gain ACT Composite score) corresponds to a change of about 0.14 ACT Composite score points that is comparable to approximately 1 month of instruction. Although coefficients of 0.05 are small, we feel they are still worthy of discussion given that many predictor variables are included, and few are likely to have large effects. Overall, the model explained 19% of the variance in academic growth (Multiple R = .434). The dependent variable is a function of ACT Composite scores with known standard error of measurement, and we estimate that an upper bound for the model R² is 77%.² Therefore, the model explained 25% of the explainable variance in academic growth.

Table 4.

Predictors of Academic Growth.

Variable	All cohorts (1996-2017)		Early cohorts (1996-2000)		Recent cohorts (2013-2017)
Variable	β	SE	β	SE	β	SE
Intercept	*−0.339	0.009	*−0.390	0.020	*−0.184	0.021
School type
Catholic	*0.131	0.006	*0.097	0.013	*0.076	0.012
Home	*−0.246	0.024	*−0.223	0.109	*−0.345	0.036
Private	*0.068	0.006	0.008	0.014	0.022	0.012
#Public, <20% FRL
Public, 20%-40% FRL	*−0.073	0.004	*−0.066	0.008	*−0.146	0.008
Public, 40%-60% FRL	*−0.174	0.004	*−0.168	0.011	*−0.273	0.009
Public, 60%-80% FRL	*−0.283	0.007	*−0.243	0.017	*−0.381	0.012
Public, >80% FRL	*−0.447	0.013	*−0.359	0.030	*−0.597	0.024
Other	*0.052	0.013	0.028	0.031	*−0.065	0.022
School locale
Rural	*−0.095	0.004	*−0.049	0.010	*−0.166	0.009
Town	*−0.063	0.004	*−0.033	0.010	*−0.096	0.009
Suburban	*−0.018	0.003	−0.007	0.009	*−0.037	0.007
#City
School class size	*0.044	0.002	*0.035	0.004	*0.032	0.003
Race/ethnicity
African American	*−0.363	0.007	*−0.306	0.020	*−0.397	0.012
Asian	*0.051	0.008	0.033	0.025	0.014	0.013
Hispanic	*−0.212	0.007	*−0.110	0.025	*−0.247	0.011
#White
Other	*−0.067	0.009	−0.049	0.026	*−0.099	0.014
Family income (student reported)
Low	*−0.122	0.004	*−0.110	0.010	*−0.096	0.010
Middle	*−0.064	0.004	*−0.056	0.008	*−0.054	0.007
#High
Parent education
High school or less					*−0.154	0.013
Some college					*−0.105	0.008
#Bachelor’s or higher
Male gender	*0.245	0.003	*0.246	0.007	*0.244	0.006
High school GPA	*0.206	0.002	*0.205	0.003	*0.218	0.003
High school courses
Other English	*−0.050	0.003	−0.003	0.007	*−0.037	0.008
Trigonometry	*0.113	0.003	*0.131	0.007	*0.124	0.006
Other advanced math	*0.052	0.003	*0.048	0.007	*0.039	0.006
Calculus	*0.127	0.003	*0.129	0.008	*0.115	0.006
Computer science	−0.002	0.004	0.008	0.008	0.001	0.007
Psychology	*−0.016	0.003	*−0.038	0.008	*−0.012	0.006
Geography	*−0.062	0.003	*−0.038	0.007	*−0.053	0.005
Economics	*−0.049	0.003	*−0.033	0.007	*−0.039	0.006
Other History	*−0.013	0.003	*−0.029	0.007	−0.007	0.006
Chemistry	*0.117	0.005	*0.132	0.012	*0.101	0.012
Physics	*0.087	0.003	*0.116	0.007	*0.059	0.006
Spanish	*−0.047	0.005	*−0.036	0.010	*−0.046	0.010
French	*−0.025	0.005	−0.009	0.011	0.001	0.011
German	−0.007	0.006	0.023	0.013	0.006	0.015
Other foreign language	*0.027	0.005	*0.048	0.012	*0.021	0.010
Art	*−0.026	0.003	*−0.035	0.007	*−0.015	0.006
Drama	*−0.027	0.004	*−0.028	0.009	*−0.024	0.008
Music	*−0.014	0.004	−0.013	0.010	0.000	0.008
AP/accelerated/honors courses
English	*0.015	0.004	0.002	0.008	*0.031	0.009
Mathematics	*0.108	0.004	*0.094	0.008	*0.115	0.009
Social studies	*0.076	0.003	*0.054	0.008	*0.062	0.008
Natural sciences	*0.056	0.004	*0.025	0.008	*0.074	0.008
Foreign language	*0.031	0.003	0.010	0.008	*0.034	0.006
ACT Interest Inventory scores
Science and Technology	*0.070	0.002	*0.054	0.004	*0.066	0.003
Arts	*0.014	0.002	*0.028	0.004	*0.008	0.003
Social Service	*−0.021	0.002	*−0.025	0.004	*−0.022	0.004
Administration and Sales	*−0.027	0.002	*−0.025	0.005	*−0.014	0.004
Business Operations	*0.042	0.002	*0.034	0.004	*0.039	0.004
Technical	*−0.037	0.002	*−0.023	0.004	*−0.039	0.004
Extracurricular activities
Instrumental music	*0.047	0.004	*0.038	0.009	*0.038	0.008
Vocal music	*−0.013	0.004	−0.004	0.009	*−0.027	0.008
Student government	0.006	0.003	−0.002	0.008	*0.027	0.007
Publications	*−0.012	0.004	0.006	0.008	−0.014	0.008
Debate	*0.034	0.005	0.020	0.011	*0.019	0.009
Dramatics, theater	*0.018	0.004	*0.041	0.008	−0.009	0.008
Religious organizations	*0.021	0.003	*0.032	0.007	0.004	0.006
Racial or ethnic organizations	*0.020	0.007	−0.002	0.018	*0.038	0.014
Varsity athletics	*−0.038	0.003	*−0.038	0.007	*−0.051	0.006
Political organizations	*0.013	0.005	−0.019	0.012	*0.035	0.011
Radio-TV	*−0.056	0.007	*−0.041	0.014	*−0.057	0.016
Fraternity, sorority, other social clubs	*−0.057	0.004	*−0.093	0.010	*−0.034	0.008
School or community service org.	*0.048	0.003	*0.041	0.008	*0.062	0.007
Model multiple R	.434		.427		.448

Note. FRL = free or reduced-price lunch; SE = standard error. β values >.05 or <−.05 are considered substantively significant. #Indicates reference group.

p < .01.

Relative to white students, African American (β = −0.363), Hispanic (β = −0.212), and students of other minority groups (β = −0.067) had lower academic growth. Low-income (β = −0.122) and middle-income (β = −0.064) students had lower growth than high-income students, and males (β = 0.245) had higher growth than females. Because the growth score was standardized, this means that males had higher average growth by 0.245 SD units, and this corresponds to a difference of about 0.66 points on the ACT Composite score scale. Because multiple linear regression was used, all coefficients should be interpreted as the expected change in the dependent variable associated with a unit change in the predictor, when holding all other predictors constant.

Relative to students attending low-poverty public high schools, students attending Catholic (β = 0.131) and private (β = 0.068) schools had more growth. Lower growth was observed in public schools with higher poverty concentrations. Relative to low-poverty public high schools, growth was 0.447 SD lower at high-poverty public schools. Lower growth was also observed for home-schooled students (β = −0.246). Relative to students attending school in cities, students attending school in rural (β = −0.095) and town (β = −0.063) locales had lower growth. Higher high school GPA was related to higher growth (β = 0.206).

The elective high school course with the strongest positive relationships with academic growth was calculus (β = 0.127), followed by chemistry, trigonometry, physics, other advanced math, and other foreign language. Social studies courses (geography, psychology, economics, and other history) and courses in the arts had negative relationships. Students taking advanced coursework (Advanced Placement, accelerated, or honors courses) had higher growth than peers not taking advanced coursework. The coefficient was largest for advanced mathematics (β = 0.108), followed by social studies (β = 0.076), and natural sciences (β = 0.056). Higher science and technology vocational interest scores (corresponding to the Investigative personality type) were related to higher growth (β = 0.070).

Relative to the other predictors, extracurricular activities were not as predictive of academic growth. Activities negatively related to academic growth included fraternity, sorority, or other social clubs (β = −0.057) and radio/TV (β = −0.056).

To summarize the results for RQ3, Figure 4 shows the 10 variables that had the largest positive beta coefficients, indicating variables related to higher academic growth. Figure 5 shows the 10 variables that had the largest negative beta coefficients, indicating variables related to lower academic growth.

Figure 4.

Variables with largest positive beta weights for predicting academic growth.

Figure 5.

Variables with largest negative beta weights for predicting academic growth.

Changes in Predictors Over the Past Two Decades

By addressing RQ3, we found that several variables were predictive of academic growth among academically advanced youth over the entire study period (1996-2017). However, it is possible that the prediction model shifted over the 22 years and that some predictors have become more (or less) important over time. To test this, we fit the regression model for the five earliest cohorts (1996-2000) and the five most recent cohorts (2013-2017) and contrasted the results. We consider predictors whose regression coefficients changed by 0.05 or more as substantively meaningful and most worthy of discussion.

The regression results are presented in Table 4 for the early and recent cohorts. The early-cohorts model accounted for 24% of the explainable variance in academic growth, and the recent-cohorts model explained 26%. The negative relationship of home-schooling and academic growth became more pronounced over time (β = −0.223 for early cohorts vs. −0.345 for recent cohorts). Similarly, the negative coefficient of higher school poverty concentration became more pronounced over time. For example, the difference between the highest and lowest public school poverty levels was 0.597 SD for the recent cohorts but only 0.359 SD for the early cohorts. Negative coefficients of rural and town school locales also became more pronounced over time.

The negative relationship of Hispanic ethnicity and academic growth became more pronounced over time (β = −0.110 for early cohorts vs. −0.247 for recent cohorts). Positive coefficients of chemistry and physics courses decreased over time. Positive coefficients of advanced courses increased over time, particularly for natural sciences.

For the recent cohorts (2011-2017), we examined the relationship of parent education level and academic growth, as these data were collected only for those cohorts. Relative to students whose parent had a bachelor’s degree or higher, those whose parents did not attend college (β = −0.154) or who had some college less than a bachelor’s degree (β = −0.105) had lower growth.

Discussion

Academic Growth Has Improved in the Past 13 Years

Academically advanced students showed a decline in growth from 1996 to 2004 but since 2005 have been on an upward trajectory, with the highest growth occurring for the students who completed high school in 2016. The level of improvement in academic growth represents a “small” effect size of 0.13 and is comparable to students in 2017 having received an extra 0.37 years of school over what students in 1996 had. First, this shows that the pattern of academic growth in this population is not static. It is possible that societal factors, such as changes in education policy or funding may have had a relationship with the growth of academically advanced students. Additionally, Benbow and Stanley (1996) noted that government K-12 gifted education funding was 0.0002% of the education budget, and Wai and Worrell (2016) showed that this remained unchanged two decades later. Although funding did not improve, academic growth did (starting in 2005), suggesting that funding may not be a core cause of improved growth. There may have been other education policies enacted that negatively or positively influenced gifted students.

Overall, the variables included in the model—sociodemographic variables, high school characteristics, high school coursework and GPA, advanced high school coursework, vocational interests, and extracurricular activities—accounted for a quarter of the explainable variance in academic growth. There are individual differences in academic growth that are not explained by the model. Aspects of cognitive development, mental health, motivation, home environment, and parental influence are examples of constructs that were not directly measured by our study but may influence academic growth. Quality and intensity of instruction, relationships with school personnel, school safety climate, and peer interactions are examples of school contextual effects also not directly measured but plausibly related to the academic growth of academically advanced students.

Academic Growth Has Not Uniformly Improved Across All Subgroups

Whereas academically advanced students overall increased in academic growth, not all subgroups of interest showed similar levels of improvement. Within each class of variables related to academic growth, we now discuss trends across the past 22 years.

Sociodemographic Variables and High School Characteristics: Disadvantaged Students Have Shown Lower Growth and Less Improvement Over Time

Type of school

Students in Catholic and private high schools tended to have the highest growth, followed by low poverty public high schools, home schooling, and finally high poverty public schools (>60% receiving free or reduced-price lunch). This suggests that, at least for academically advanced students, Catholic and private are schooling environments in which academically advanced students have higher performance relative to public school environments or homeschooling environments. We caution the reader, however, that our study is not a causal design and students who attend Catholic or other specific types of schools (e.g., homeschooling) may be inherently different from students who do not. Some research on general population students has documented positive Catholic school effects (e.g., Wenglinsky, 2007; West & Woessmann, 2010). Much research suggests that for students in the general population, students in private schools have higher achievement than students in public schools, even after controlling for multiple student and school characteristics (e.g., Braun, Jenkins, & Grigg, 2006; Coleman, Hoffer, & Kilgore, 1982; Petersen & Llaudet, 2006); however, some scholars have noted that differential effects of public and private schools are unclear (e.g., Wenglinsky, 2007), with other scholars documenting advantages of public schools (e.g., Lubienski & Lubienski, 2013). Our analysis shows that, for whatever reason, academically advanced students in Catholic and private schools tended to have higher growth than their public school counterparts.

Similar to private and public school debates, there is disagreement on how homeschooling overall influences general population students (e.g., Jones & Gloeckner, 2004; Rudner, 1999; Kunzman & Gaither, 2013; Snyder, 2013; Welner & Welner, 1999). The finding that academically advanced students who attend high school at home showed lower growth may be due to such environments not being as consistent or standardized as other environments or other factors. Of course, it is important to note that there is great variability across each of these schooling environments and that these conclusions are based on averages across these types of schools.

Students with disabilities showed improvement

Although students with disabilities demonstrated less growth than the total group, they still showed improvement in growth from 2005 to 2017. Some students received special testing accommodations, and it is possible that the improvement in growth is partly explained by changes in the quality of accommodations over the 13-year period. The lower growth from Grade 7 to Grade 11/12 among students with disabilities is consistent with suggestions that gifted students who also have disabilities, often known as “twice exceptional,” are not receiving adequate attention in terms of academic growth (Assouline & Whiteman, 2011; Brody & Mills, 1997; Reis, Baum, & Burke, 2014). However, over time they have largely closed the gap with the total group.

Groups with relatively high growth

Relative to the general population, Asians were underrepresented and Whites overrepresented in the study sample, something typically found in other gifted samples (Lubinski, Webb, Morelock, & Benbow, 2001; Yoon & Gentry, 2009). Asians showed higher academic growth than Whites. Males had higher growth than females, suggesting that talent development opportunities or potential barriers may still be influencing academically advanced females (e.g., Reis, 2002).

Groups with less improvement in growth

Low-income students demonstrated lower growth relative to the total group and no improvement in growth over the 22-year study period. This aligns with the literature indicating that gifted low-income students are disadvantaged and losing ground compared with their advantaged counterparts (e.g., Loveless, 2016; Plucker & Peters, 2016; Wyner et al., 2007). Students whose parent(s) do not have a bachelor’s degree or higher tended to have lower growth than their counterparts. This collectively suggests that there are widening gaps between the more advantaged and less advantaged academically advanced students (Plucker & Peters, 2016; Wai & Worrell, 2016) that likely has long-term implications not only for college admissions and readiness (Bastedo & Jaquette, 2011) but also for long-term educational and occupational achievement among high-level careers (Park et al., 2007; Wai et al., 2005).

Elective High School Coursework and GPA: Science, Technology, Engineering, and Mathematics Courses Are Associated With Higher Academic Growth

The types of courses students elect to take are in part a reflection of their personal interests as well as availability and may serve as an intermediate form of advanced coursework or educational enrichment. However, in some cases it is assumed that students took rigorous courses to be “college ready.” The finding that higher growth was associated largely with science, technology, engineering, and mathematics (STEM) courses (i.e., physics, trigonometry, chemistry, calculus) suggests that STEM coursework may be related to academic growth. Overall, students with a higher high school GPA also had higher growth than their peers. This could be due to these students being more serious about the grades they earn, indicating a stronger work ethic. It could also be that both higher GPA and higher growth are caused by other similar sets of factors, such as work ethic and study habits. This is consistent with research suggesting that high school GPA is a positive predictor of college performance broadly (Kobrin et al., 2008; Sackett et al., 2012) and with other research showing positive relationships between grades and ACT score growth in high school students (Sawyer, 2008).

Advanced High School Coursework

Overall, students who took advanced coursework (AP, accelerated, or honors courses) had significantly higher academic growth. This was strongest for advanced mathematics, social studies, and natural sciences, again suggesting that STEM coursework is associated with enhanced academic growth. This links with research showing that a high dosage of K-12 STEM course work is predictive of long-term STEM outcomes (Wai, Lubinski, Benbow, & Steiger, 2010). These findings also align with a large body of research literature supporting academic acceleration and enrichment as effective interventions for academically advanced students (Assouline et al., 2015; Hertzog & Chung, 2015; Schiel, 1998; Steenbergen-Hu, Makel, & Olszewski-Kubilius, 2016) showing that more participation in advanced coursework is associated with improved outcomes for gifted students. These findings also align with research within gifted populations specifically linking participation in AP courses with better long-term outcomes (Bleske-Rechek et al., 2004).

Vocational Interests

Academically advanced students who had higher Investigative scores, which indicate an interest in understanding natural phenomena in the natural sciences, had higher academic growth. In contrast, students who had higher Realistic scores, those with more hands-on mechanical and spatial interests, had lower academic growth, which may be explained in part by the fact that the ACT includes primarily math and verbal, but not spatial, measures. Spatial growth, if present, would not be captured (Webb, Lubinski, & Benbow, 2007). This may also indicate in part that students with higher realistic scores tend to have interests that lie in hands-on activities, something that an academic assessment such as the ACT does not directly assess (Gohm, Humphreys, & Yao, 1998; see Nye, Su, Rounds, & Drasgow, 2012 for a 60-plus year summary of vocational interests and performance). It appears that having more STEM-related interests, coupled with organization and procedural interests, is related to academic growth broadly.

Extracurricular Activities

Overall, extracurricular activities were not as predictive of academic growth as the other groups of variables included in the model, suggesting that what students do outside of the classroom does not have much of a collective relationship with academic development. Research generally has found positive associations between extracurricular activity participation and academic achievement (Broh, 2002; Cooper et al., 1999; Marsh & Kleitman, 2002). In this study, participating in school or community service organizations, instrumental music, debate, religious organizations, or racial or ethnic organizations was related to higher academic growth. Perhaps participating in community service or racial/ethnic organizations shows values of service and diversity that may be linked to growth. The practice of debate may be linked to academic growth, in part, due to learning to construct logical arguments. The finding that participation in social clubs and radio/TV was negatively related to academic growth suggests that higher involvement in nonacademic activities (perhaps at the expense of academic activities) may be associated with lower academic growth. More broadly, it suggests that how academically talented students allocate their time is of potential importance (e.g., Makel et al., 2015). Though extracurriculars overall were not predictive of academic growth, it is possible they have a relationship with factors other than academic growth, such as noncognitive skills or other forms of development (e.g., Heckman, Stixrud, & Urzua, 2006) for talented students.

Which Predictors of Academic Growth Have Become More Important Over Time?

In the prior sections we discussed variables and how they predicted academic growth. Overall, there was not much change in the predictive model across time. High school poverty, rural and town school locales, Hispanic ethnicity, and home schooling were all associated with lower growth, and this became more pronounced over time. This suggests that poor students in rural areas of Hispanic ethnicity, even in comparison to disadvantaged students generally, are losing ground in recent years. While some elective STEM coursework, which tended to have a positive relationship with growth, showed decreased strength over time, the positive coefficients of advanced courses, particularly in the natural sciences, increased over the 22-year period.

Potential Areas for Educational Intervention Based on These Findings

We used an exploratory modeling approach, including as many variables as possible to potentially uncover factors predictive of academic growth that might be leveraged in some capacity through educational interventions. Some of these variables likely are malleable, and we provide in this section some potential directions for educational intervention research and application. However, we caution the reader that our study is correlational, not causal.

Geographic Location and Poverty

Providing greater access to programs for advanced youth in rural areas and high-poverty high schools may benefit their academic growth. These environments, and perhaps other factors surrounding them, would be worth investigating further (e.g., Azano, Callahan, Missett, & Brunner, 2014; Puryear & Kettler, 2017).

School Type

Future research might examine what, if any, elements of a Catholic and/or private school may be associated with higher academic growth: is it the quality of education, peer effects, teacher effects, or a combination of these or other factors? If uncovered, replicating elements of such environments in public schools could be helpful. Although home schools may not be the best environment for academic growth of academically advanced students, there may be other benefits, or ways such educational environments could be improved (e.g., Witham, 1997).

STEM Coursework

A thread that ran through this study is the consistent correlation of STEM courses with academic growth broadly. Therefore, STEM education may actually enhance academic growth not just in STEM areas such as math and science but also overall. This may broaden the discussion surrounding the importance of STEM education, not just for STEM outcomes (e.g., Wai et al., 2010) but broad outcomes (e.g., Lubinski, Benbow, & Kell, 2014).

Advanced Coursework

A large body of literature already supports the idea that advanced coursework in the form of academic acceleration and enrichment is related to enhancing the talent of academically advanced students (e.g., Assouline et al., 2015; Bleske-Rechek et al., 2004; Wai et al. 2010). This study confirms the importance of such interventions and illustrates that advanced coursework may be a ready intervention for students from disadvantaged backgrounds that therefore should be more widely available and easily accessible (e.g., see Kolluri, 2018).

Investigative and Conventional Interests

Investigative and Conventional interests were associated with higher growth. Future study of the relationships between interests, course-taking, and achievement growth may be worthwhile to pursue. For example, consideration should be given to whether developing STEM and organizational-related interests in this population would be at the expense of developing interests in areas such as the arts and social services (e.g., Zakaria, 2016).

Extracurriculars

Involvement in social clubs, radio/TV, and varsity athletics was related to lower academic growth, while involvement in school and community service, instrumental music, debate, religious organizations, and racial or ethnic organizations was related to higher academic growth. Extracurricular choice typically is not made based on consideration of academic benefits but on other considerations (e.g., time, enjoyment, social opportunities, fitness); however, parents and students should be aware that investment of time in one area of development correspondingly means less investment in another, likely influencing areas of differential growth (e.g., see Makel et al., 2015).

Limitations and Future Research Directions

One important limitation is the issue of access to the talent search. Our sample included students who had such access, but it is likely that students who are identified this early on in their educational trajectories to be a part of the talent search are already select. For example, Table 1 shows that when comparing poverty level of schools, the talent search sample is more financially secure than the general population. This means there are more academically advanced students at these schools and/or they are more likely to participate in the talent search. This is a problem only if the two phenomena examined in this study (growth trends over the 22-year period, and predictors of academic growth) are different for high- and low-poverty schools. Furthermore, students in the sample resided mostly in the Midwest and South United States, with very little representation from the Northeast and West regions. Again, this is only a problem if the studied phenomena actually vary by geographic region. Additional research is needed to determine the extent that findings vary by school poverty level and geographic region.

The study is based on a large sample of self-selected youth who participated in a talent search program in 7th grade and later took the ACT test in high school. Because the sample composition changed extensively over the 22-year period, the growth trend could be due, in part, to changes in the U.S. population, as well as changes in who participated in the talent search and the ACT test. For example, over the 22 years, the sample trended toward becoming more Hispanic and less White, more from suburban and urban locales and less from rural and town locales, and more from higher poverty schools and less from lower poverty schools. Some of these changes (e.g., more students from suburban and urban locales) are associated with higher academic growth, while some are associated with lower academic growth (e.g., more Hispanic students and more from higher poverty schools). Moreover, the changes generally coincide with shifting demographics and population trends in the United States. Additional analyses could inform how well the study generalizes to all academically advanced youth by examining predictors of talent search participation, and if and how those predictors have changed over time. For example, among all ACT-tested high school students, what variables predict participation in talent search? Such analyses could also inform equity and access issues for the talent search.

The ACT test is only one measure of academic achievement and additional research could examine whether these study findings replicate using alternative measures. We examined the ACT Composite score, which is the average across mathematics, science, reading, and English. This is an important first step in investigating growth generally. The study is strengthened by using a highly reliable measure at two time points. Future research could examine whether the prediction models vary across subject areas. The study could also be replicated using SAT test scores at Grade 7 and during high school, and this would partly address the geographic limitation of the study. Perhaps more important, reliance on ACT test scores limits the construct generalizability of this study to general academic achievement and college readiness. There are many other constructs that are important for the development of academically advanced youth, including personality dimensions such as conscientiousness, teamwork, creativity and other factors that are not directly addressed in the present study.

We included as many academic, extracurricular, sociodemographic, and other potential intervening variables that had the potential to influence academic growth in our model for which we had sufficient samples and that made sense to include. However, we did not have information on every influential variable possible, and much of the variance in academic growth remained unexplained. Despite this limitation, our sample size was quite large—especially for a sample of academically advanced youth—and for the variables studied, these findings are likely robust. Of course, some of our variables like high school coursework and grades are self-reported and thus also are limited in that sense.

Although we uncovered several predictors of academic growth, this analysis focused on the factors that predicted growth in the sample as a whole. Future research should be conducted to examine which of these factors explains growth among disadvantaged students and other subgroups to better understand intervention points that could be leveraged to help the talent development of these students. Current limits on talent development programs may be constrained by the schools and communities they include; thus, the purposeful inclusion of historically underserved populations might improve and inform our understanding of talent development.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Jonathan Wai

Notes

Author Biographies

Jonathan Wai is assistant professor of education policy and psychology, the 21st Century Endowed Chair in Education Policy at the University of Arkansas in the Department of Education Reform, and also holds a joint (courtesy) appointment in the Department of Psychology. His research examines education policy through the lens of psychology, in particular how individual and contextual factors collectively contribute to and take away from talent development and the eventual acquisition of expertise.

Jeff Allen is a statistician and director in validity and efficacy research at ACT. He specializes in longitudinal research of educational outcomes, student growth models, and validation of college readiness measures.

References

ACT. (2009). ACT Interest Inventory Technical Manual. Iowa City, IA: Author.

ACT. (2016). The ACT profile report: National, graduating class 2016. Iowa City, IA: Author.

ACT. (2017). The ACT Technical Manual. Iowa City, IA: Author.

Allen

(2016). How do 7th graders who take the ACT test score as 11th or 12th graders? ACT Technical Brief. Iowa City, IA: ACT.

Assouline

S. G.

Colangelo

VanTassel-Baska

Lupkowski-Shoplik

A. E.

(Eds.). (2015). A nation empowered: Evidence trumps the excuses that hold back America’s brightest students. Iowa City, IA: Belin-Blank Center for Gifted and Talented Education.

Assouline

S. G.

Lupkowski-Shoplik

(2012). The talent search model of gifted identification. Journal of Psychoeducational Assessment, 30, 45-59. doi:10.1177/0734282911433946

Assouline

S. G.

Whiteman

C. S.

(2011). Twice-exceptionality: Implications for school psychologists in the post–IDEA 2004 era. Journal of Applied School Psychology, 27, 380-402. doi:10.1080/15377903.2011.616576

Augustine

(2005). Rising above the gathering storm: Energizing and employing America for a brighter economic future. Washington, DC: National Academies Press.

Azano

A. P.

Callahan

C. M.

Missett

T. C.

Brunner

(2014). Understanding the experiences of gifted education teachers and fidelity of implementation in rural schools. Journal of Advanced Academics, 25, 88-100. doi:10.1177/1932202X14524405

10.

Bastedo

M. N.

Jaquette

(2011). Running in place: Low income students and the dynamics of higher education stratification. Educational Evaluation and Policy Analysis, 33, 318-339. doi:10.3102/0162373711406718

11.

Benbow

C. P.

Stanley

J. C.

(1996). Inequity in equity: How “equity” can lead to inequity for high-potential students. Psychology, Public Policy, and Law, 2, 249-292. doi:doi:10.1037/1076-8971.2.2.249

12.

Benjamin

(2018). Factors influencing learning. In Biswas-Diener

Diener

(Eds.), Noba textbook series: Psychology. Champaign, IL: DEF. Retrieved from https://nobaproject.com/modules/factors-influencing-learning

13.

Betebenner

D. W.

(2009). Norm- and criterion-referenced student growth. Educational Measurement: Issues and Practice, 28, 42-51. doi:10.1111/j.1745-3992.2009.00161.x

14.

Bleske-Rechek

Lubinski

Benbow

C. P.

(2004). Meeting the educational needs of special populations. Psychological Science, 15, 217-224. doi:10.1111/j.0956-7976.2004.00655.x

15.

Braun

Jenkins

Grigg

(2006). Comparing private schools and public schools using hierarchical linear modeling (U.S. Department of Education NCES 2006-461). Retrieved from https://nces.ed.gov/nationsreportcard/pdf/studies/2006461.pdf

16.

Brody

L. E.

Mills

C. J.

(1997). Gifted children with learning disabilities: A review of the issues. Journal of Learning Disabilities, 30, 282-296. doi:10.1177/002221949703000304

17.

Broh

B. A.

(2002). Linking extracurricular programming to academic achievement: Who benefits and why? Sociology of Education, 75, 69-95. doi:10.2307/3090254

18.

Castellano

K. E.

A. D.

(2013). A practitioner’s guide to growth models. Washington, DC: Council of Chief State School Officers.

19.

Castellano

K. E.

A. D.

(2015). Practical differences among aggregate-level conditional status metrics: From median student growth percentile to value-added models. Journal of Educational and Behavioral Statistics, 40, 35-68. doi:10.3102/1076998614548485

20.

Coleman

J. S.

Hoffer

Kilgore

(1982). High school achievement. New York, NY: Basic Books.

21.

Cohen

(1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Lawrence Erlbaum.

22.

College Board. (2016). About the SAT. Retrieved from https://collegereadiness.collegeboard.org/sat

23.

Cooper

Valentine

J. C.

Nye

Lindsay

J. J.

(1999). Relationships between five after-school activities and academic achievement. Journal of Educational Psychology, 91, 369-378. doi:10.1037/0022-0663.91.2.369

24.

Dynarski

S. M.

(2018). ACT/SAT for all: A cheap, effective way to narrow income gaps in college. Retrieved from https://www.brookings.edu/research/act-sat-for-all-a-cheap-effective-way-to-narrow-income-gaps-in-college/

25.

Executive Office of the President of the United States. (2016). Artificial intelligence, automation, and the economy. Retrieved from https://obamawhitehouse.archives.gov/sites/whitehouse.gov/files/documents/Artificial-Intelligence-Automation-Economy.PDF

26.

Glander

(2016). Documentation to the 2014-15 Common Core of Data (CCD) Universe Files (NCES 2016-077). Washington, DC: National Center for Education Statistics, U.S. Department of Education. Retrieved from http://nces.ed.gov/pubsearch/

27.

Gohm

C. L.

Humphreys

L. G.

Yao

(1998). Underachievement among spatially gifted students. American Educational Research Journal, 35, 515-531. doi:doi:10.3102/00028312035003515

28.

Hanushek

E. A.

Woessmann

(2015). The knowledge capital of nations: Education and the economics of growth. Cambridge: MIT Press.

29.

Heckman

(2000). Policies to foster human capital. Research in Economics, 54, 3-56.

30.

Heckman

J. J.

Stixrud

Urzua

(2006). The effects of cognitive and noncognitive abilities on labor market outcomes and social behaviors (Working Paper No. 12006). Cambridge, MA: National Bureau of Economic Research.

31.

Hertzog

N. B.

Chung

R. U.

(2015). Outcomes for students on the fast track to college: Early college entrance programs at the University of Washington. Roeper Review, 37, 39-49. doi:10.1080/02783193.2014.976324

32.

Holland

J. L.

(1997). Making vocational choices: A theory of vocational personalities and work environments (3rd ed.). Odessa, FL: Psychological Assessment Resources.

33.

Hyman

(2017). ACT for all: The effect of mandatory college entrance exams on postsecondary attainment and choice. Education Finance and Policy, 12, 281-311. doi:10.1162/EDFP_a_00206

34.

Jones

Gloeckner

(2004). First-year college performance: A study of home school graduates and traditional school graduates. Journal of College Admission, Spring, 17-20. Retrieved from https://files.eric.ed.gov/fulltext/EJ682484.pdf

35.

Kobrin

J. L

Patterson

B. F.

Shaw

E. J.

Mattern

K. D.

Barbuti

S. M.

(2008). Validity of the SAT for predicting first-year college grade point average (College Board Research Report No. 2008-5). New York, NY: College Board.

36.

Kolluri

(2018). Advanced Placement: The dual challenge of equal access and effectiveness. Review of Educational Research, 88, 671-711. doi:10.3102/0034654318787268

37.

Kuncel

N. R.

Hezlett

S. A.

Ones

D. S.

(2004). Academic performance, career potential, creativity, and job performance: Can one construct predict them all? Journal of Personality and Social Psychology, 86, 148-161. doi:10.1037/0022-3514.86.1.148

38.

Kunzman

Gaither

(2013). Homeschooling: A comprehensive survey of the research. Other Education: The Journal of Educational Alternatives, 2, 4-59.

39.

Lewin

(2013). Where the SAT and ACT dominate. New York Times. Retrieved from https://archive.nytimes.com/www.nytimes.com/interactive/2013/08/04/education/edlife/where-the-sat-and-act-dominate.html

40.

Loveless

(2016). How well are American students learning? The 2016 Brown Center Report on American Education. Washington, DC: Brookings Institution. Retrieved from https://www.brookings.edu/wp-content/uploads/2016/03/Brown-Center-Report-2016.pdf

41.

Lubienski

C. A.

Lubienski

S. T.

(2013). The public school advantage: Why public schools outperform private schools. Chicago, IL: University of Chicago Press.

42.

Lubinski

Benbow

C. P.

(2006). Study of mathematically precocious youth after 35 years: Uncovering antecedents for the development of math-science expertise. Perspectives on Psychological Science, 1, 316-345. doi:10.1111/j.1745-6916.2006.00019.x

43.

Lubinski

Benbow

C. P.

Kell

H. J.

(2014). Life paths and accomplishments of mathematically precocious males and females four decades later. Psychological Science, 24, 2217-2232. doi:10.1177/0956797614551371

44.

Lubinski

Benbow

C. P.

Ryan

(1995). Stability of vocational interests among the intellectually gifted from adolescence to adulthood: A 15-year longitudinal study. Journal of Applied Psychology, 80, 196-200. doi:10.1037/0021-9010.80.1.196

45.

Lubinski

Webb

R. M.

Morelock

M. J.

Benbow

C. P.

(2001). Top 1 in 10,000: A 10-year follow up of the profoundly gifted. Journal of Applied Psychology, 86, 718-729. doi:10.1037/0021-9010.86.4.718

46.

Makel

M. C.

Wai

Putallaz

Malone

P. S.

(2015). The academic gap: An international comparison of the time allocation of academically talented students. Gifted Child Quarterly, 59, 177-189. doi:10.1177/0016986215578746

47.

Marsh

H. W.

Kleitman

(2002). Extracurricular school activities: The good, the bad, and the non-linear. Harvard Educational Review, 72, 464-514. doi:10.17763/haer.72.4.051388703v7v7736

48.

National Science Technology Council. (2016). The national artificial intelligence research and development strategic plan. Executive office of the President of the United States. Retrieved from https://www.nitrd.gov/PUBS/national_ai_rd_strategic_plan.pdf

49.

The Networking and Information Technology Research and Development Program. (2016). The federal big data research and development strategic plan. Retrieved from http://digitalcommons.unl.edu/cgi/viewcontent.cgi?article=1019&context=scholcom

50.

Nye

C. D.

Rounds

Drasgow

(2012). Vocational interests and performance: A quantitative summary of over 60 years of research. Perspectives on Psychological Science, 7, 384-403. doi:10.1177/1745691612449021

51.

Olszewski-Kubilius

(1998). Talent search: Purposes, rationale, and role in gifted education. Journal of Advanced Academics, 9, 106-113. doi:10.1177/1932202X9800900303

52.

Park

Lubinski

Benbow

C. P.

(2007). Contrasting intellectual patterns predict creativity in the arts and sciences. Psychological Science, 18, 948-952. doi:10.1111/j.1467-9280.2007.02007.x

53.

Petersen

P. E.

Llaudet

(2006, August). On the public-private school achievement debate. Retrieved from http://scholar.harvard.edu/files/ellaudet/files/on_the_public-private_school_achievement_debate.pdf

54.

Plucker

J. A.

Peters

S. J.

(2016). Excellence gaps in education: Expanding opportunities for talented students. Cambridge, MA: Harvard Education Press.

55.

Puryear

J. S.

Kettler

(2017). Rural gifted education and the effect of proximity. Gifted Child Quarterly, 61, 143-152. doi:10.1177/0016986217690229

56.

Reis

S. M.

(2002). Internal barriers, personal issues, and decisions faced by gifted and talented females. Gifted Child Today, Winter, 1-21. Retrieved from https://gifted.uconn.edu/schoolwide-enrichment-model/internal_barriers_gifted_females/

57.

Reis

S. M.

Baum

S. M.

Burke

(2014). An operational definition of twice-exceptional learners: Implications and applications. Gifted Child Quarterly, 58, 217-230. doi:10.1177/0016986214534976

58.

Rindermann

Thompson

(2011). Cognitive capitalism: The effect of cognitive ability on wealth, as mediated through scientific achievement and economic freedom. Psychological Science, 22, 754-763. doi:10.1177/0956797611407207

59.

Rubin

D. B.

(1996). Multiple Imputation after 18+ Years. Journal of the American Statistical Association, 91, 473-489.

60.

Rudner

L. M.

(1999). Scholastic achievement and demographic characteristics of home school students in 1998. Education Policy Analysis Archives, 7, 8. doi:doi:10.14507/epaa.v7n8.1999

61.

Sackett

P. R.

Kuncel

N. R.

Beatty

A. S.

Rigdon

J. L.

Shen

Kiger

T. B.

(2012). The role of socioeconomic status in SAT-grade relationships and in college admissions decisions. Psychological Science, 23, 1000-1007. doi:10.1177/0956797612438732

62.

Sanchez

Buddin

(2016). How accurate are self-reported high school courses, course grades, and grade point averages? (ACT Research Report 2016-3). Iowa City, IA: ACT.

63.

Sawyer

(2008). Benefits of additional high school course work and improved course performance in preparing students for college (ACT Research Report 2008-1). Iowa City, IA: ACT.

64.

Schiel

(1998). Academic benefits in high school of an intensive summer program for academically talented seventh graders (ACT Research Report 1998-4). Iowa City, IA: ACT.

65.

Schmitt

Keeney

Oswald

F. L.

Pleskac

T. J.

Billington

A. Q.

Sinha

Zorzie

(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, 1479-1497. doi:10.1037/a0016810

66.

Snyder

(2013). An evaluative study of the academic achievement of homeschooled students versus traditionally schooled students attending a Catholic university. Catholic Education: A Journal of Inquiry and Practice, 16, 288-308. doi:10.15365/joce.1602042013

67.

Steenbergen-Hu

Makel

M. C.

Olszewski-Kubilius

(2016). What one hundred years of research says about the effects of ability grouping and acceleration on K-12 students’ academic achievement: Findings of two second-order meta-analyses. Review of Educational Research, 86, 849-899. doi:10.3102/0034654316675417

68.

Rounds

(2015). All STEM fields are not created equal: People and things interests explain gender disparities across STEM fields. Frontiers in Psychology: Developmental, 6, 189. doi:10.3389/fpsyg.2015.00189

69.

Subotnik

R. F.

Olszewski-Kubilius

Worrell

F. C.

(2011). Rethinking giftedness and gifted education: A proposed direction forward based on psychological science. Psychological Science in the Public Interest, 12, 3-54. doi:10.1177/1529100611418056

70.

Wai

Lubinski

Benbow

C. P.

(2005). Creativity and occupational accomplishments among intellectually precocious youths: An age 13 to age 33 longitudinal study. Journal of Educational Psychology, 97, 484-492. doi:10.1037/0022-0663.97.3.484

71.

Wai

Lubinski

Benbow

C. P.

Steiger

J. H.

(2010). Accomplishment in science technology, engineering, and mathematics (STEM) and its relation to STEM educational dose: A 25-year longitudinal study. Journal of Educational Psychology, 102, 860-871. doi:10.1037/a0019454

72.

Wai

Worrell

F. C.

(2016). Helping disadvantaged and spatially talented students fulfill their potential: Related and neglected national resources. Policy Insights from the Behavioral and Brain Sciences, 3, 122-128. doi:10.1177/2372732215621310

73.

Webb

R. M.

Lubinski

Benbow

C. P.

(2007). Spatial ability: A neglected dimension in talent searches for intellectually precocious youth. Journal of Educational Psychology, 99, 397-420. doi:10.1037/0022-0663.99.2.397

74.

Welner

K. G.

(1999). Contextualizing homeschooling data: A response to Rudner. Education Policy Analysis Archives, 7. doi:10.14507/epaa.v7n13.1999

75.

Wenglinsky

(2007). Are private high schools better academically than public high schools? Washington, DC: Center on Education Policy.

76.

West

M. R.

Woessmann

(2010). Every Catholic child in a Catholic school: Historical resistance to state schooling, contemporary private competition and student achievement across countries. Economic Journal, 120, F229-F255. doi:10.1111/j.1468-0297.2010.02375.x

77.

Witham

J. H.

(1997). Public or private schools: A dilemma for gifted students? Roeper Review, 19, 137-141. doi:10.1080/02783199709553809

78.

Wyner

J. S.

Bridgeland

J. M.

DiIulio

J. J.

(2007). Achievement trap: How America is failing millions of high-achieving students from lower-income families. Landsdowne, VA: Jack Kent Cooke Foundation.

79.

Yoon

S. Y.

Gentry

(2009). Racial and ethnic representation in gifted programs: Current status of and implications for gifted Asian American students. Gifted Child Quarterly, 43, 121-136. doi:10.1177/0016986208330564

80.

Zakaria

(2016). In defense of a liberal education. New York, NY: W. W. Norton.