Charting the Pathway to STEM: How Middle School Socialization and Science Growth Trajectories Predict Adult Career Success

Abstract

Middle school is a critical period for science education, yet the collective impact of socialization agents on students’ longitudinal science learning trajectories and subsequent STEM careers remains underexplored. This study investigates how seventh-grade (typically aged 12–13) socialization agents—parental educational encouragement, peer academic support, constructive school learning environment, and student self-esteem—collectively shape the developmental growth trajectories of science performance throughout middle school and predict the attainment of a college STEM degree and later engagement in STEM professions in adulthood. Using five-wave longitudinal data from the Longitudinal Study of American Youth (LSAY, N = 3116), we employed latent growth curve modeling (LGCM) to analyze these relationships. Results indicated that all four grade-7 socialization agents significantly predicted a higher initial level of science achievement. In addition, parental encouragement and a constructive school learning environment also predicted a positive growth rate of science achievement. Furthermore, both the initial level and growth of science performance significantly predicted successful graduation with a STEM degree. These middle school science trajectories, along with obtaining a STEM degree, sequentially mediated the relationships between the grade-7 socialization agents and adult STEM career engagement. The findings underscore the necessity of educational policies and interventions that foster a synergistic pro-learning socialization context in middle school to bolster students’ science education and pave the way for long-term STEM success.

1. Introduction

The global advancement of science, technology, engineering, and mathematics (STEM) is widely recognized as a cornerstone of economic competitiveness, technological innovation, geopolitical standing, and societal progress (Anderson & Li, 2020; Tasos et al., 2018). Consequently, understanding and fostering the pathways that lead individuals to become STEM professionals is a paramount concern for educators, policymakers, and researchers across nations alike. While significant attention has been paid to STEM education at the collegiate and high school levels, early adolescence—particularly the middle school years—represents a critically underexplored yet formative period for the development of foundational science knowledge, academic identity, and future career aspirations (Crosnoe & Johnson, 2011). During this pivotal stage, students are introduced to core scientific disciplines such as biology, physics, and chemistry, which establish the essential groundwork for all subsequent STEM learning and development (English, 2017; Kloser et al., 2018). A deficit in science learning during middle school can therefore foreseeably hinder a student’s entire STEM educational trajectory, as proficiency in these subjects is cumulative and progressive, relying heavily on earlier acquired knowledge and skills (Mau et al., 1995; Newman et al., 2015; Ocal et al., 2021).

Beyond the academics, students in early adolescence are highly susceptible to influences from their surrounding social ecology. Their educational and occupational development is not shaped in a vacuum but is profoundly affected by a constellation of socialization agents, including parents, peers, school environments, and their own intrapersonal resources (Bottia et al., 2021; Keung & Ho, 2019; Sommerfeld, 2016; Yeung & Xia, 2023). Empirical evidence suggests that parental educational encouragement, peer academic support, a constructive school learning climate, and positive student self-esteem are pivotal factors that collectively create a pro-learning context (Bornstein et al., 2018; Bottia et al., 2021; Woo et al., 2021). These agents have been independently linked to better academic performance across subjects, including science, and to stronger intentions to pursue higher education and STEM careers (Tey et al., 2020; Yeung & Xia, 2023). However, existing research often examines these factors in isolation, neglecting their collective and synergistic impact (Bottia et al., 2023; Luo et al., 2022; Mitsopoulou & Pavlatou, 2021; Sahin et al., 2018). Moreover, a significant gap remains in understanding how this collective socialization context at the beginning of middle school influences not just a static measure of science performance, but the developmental and growth trajectories of science learning across the entire middle school period. This oversight is theoretically and methodologically significant. Educational performance, especially science learning, is not a fixed attribute but a dynamic and malleable process (Arens et al., 2017; Pekrun et al., 2017). Students’ science learning trajectories—capturing both their initial proficiency (intercept) and their rate of change (slope) over time—offer a more nuanced understanding of their academic development than a single-point assessment (Arens et al., 2017; Yeung & Igarashi, 2025). These trajectories are likely a key mechanism through which early socialization influences long-term outcomes.

Grounded in the integrative frameworks of Social Cognitive Career Theory (SCCT) and Life Course Theory, this study posits that STEM development is a continuum and developmental course (Lent et al., 1994; Pallas, 2003). SCCT emphasizes that environmental and personal factors shape academic and career interests, choices, and performance (Lent & Brown, 2013; Lent et al., 1994), while Life Course Theory suggests that early life stages and achievements have enduring consequences for later outcomes (Pallas, 2003). Middle school, in this view, serves as a “pre-action planning phase” where the foundational competencies and aspirations for STEM are established, setting the stage for the “action phase” of obtaining a STEM degree in college and the “post-action pursuing phase” of entering a STEM profession (Lent et al., 1994; Lent et al., 2018). Nevertheless, no research to date has simultaneously integrated the collective influence of key socialization agents, modeled the dynamic trajectories of science learning they predict, and traced their distal effects on the tangible outcomes of earning a STEM degree and securing STEM employment in adulthood. Therefore, the present study aims to address this gap by investigating the interconnected longitudinal relationships between grade-7 (typically aged 12–13) socialization agents, the developmental and growth trajectories of students’ science performance across middle school, and their ultimate STEM attainment in adulthood. Using a nationally representative longitudinal sample and latent growth curve modeling (LGCM), this research examines how parental encouragement, peer support, school environment, and self-esteem collectively predict both the initial level and growth of science achievement. It further tests whether these science learning trajectories mediate the influence of early socialization on the likelihood of graduating with a STEM degree (Ahmed, 2018; Wright et al., 2017) and, subsequently, working in a STEM profession.

The findings of this study hold substantial implications. By elucidating the dynamic process through which a supportive middle school ecology fosters science learning and channels students into STEM careers, this research provides a robust evidence base for targeted educational reforms and policy innovations. It underscores the necessity of moving beyond siloed interventions and towards a holistic approach that mobilizes resources across family, peer, school, and individual levels to cultivate the next generation of STEM professionals, especially for those students in disadvantaged learning contexts.

2. Conceptual Framework and Theoretical Rationale for a Dynamic Approach to STEM Development

Cultivating a robust pipeline of STEM graduates and professionals is recognized as a progressive, resource-intensive endeavor critical to economic competitiveness, societal development, cultural advancement, and technological innovation at national and international levels (English, 2017; Larkin & Lowrie, 2022). This is because STEM graduates and professionals not only must master language proficiency and sociocultural awareness, similar to their peers in non-STEM fields, but also, more importantly, need competent mathematics and science knowledge and skills applicable to their STEM field (Penprase, 2020; Tasos et al., 2018). Grounding this investigation, SCCT and Life Course Theories provide a powerful integrative framework for understanding STEM development as a long-term, dynamic process rather than a singular event (Lent et al., 1994; Lent et al., 2018; Pallas, 2003). SCCT posits that individuals’ academic and career interests, choices, and performance are shaped by a triadic reciprocal relationship between environmental factors, personal attributes, and behavioral outcomes (Lent & Brown, 2013; Lent et al., 1994). Concurrently, Life Course Theory emphasizes that trajectories established in early developmental periods have profound and enduring consequences for later-life outcomes (Pallas, 2003). Synthesizing these perspectives, middle school emerges as a critical “pre-action planning phase” where foundational competencies, interests, and aspirations for STEM are established. This phase sets the stage for subsequent milestones: the “action phase” of obtaining a STEM degree in college and the “post-action pursuing phase” of entering a STEM profession (Lent et al., 2018; Pallas, 2003). Consequently, the educational experiences and socialization context of early adolescence are theorized to cast a long shadow, influencing adult career success through a chain of mediating processes.

Life Course Theory posits that early adolescence, particularly the middle school years, is a critical formative period for students to develop and cultivate science interests, academic motivation, and career plans (Gecas, 2003; Pallas, 2003); this developmental stage is hence closely related to students’ future STEM development in adulthood (English, 2017; Larkin & Lowrie, 2022; Ozulku & Kloser, 2023). Pertinently, SCCT suggests that educational achievement and career development of students in adulthood are formulated by their environmental and individual socialization factors in earlier stages (Lent & Brown, 2013; Lent et al., 1994), in which parents, peers, school, and students’ self-concept are reckoned by the theory as the most intimate and influential socialization agents to sway students’ educational and occupational development, including their science performance in middle school and STEM achievement adulthood. Conceptually, SCCT frames middle school as the ‘pre-action planning phase’ for exploring and developing educational motivations and career plans. College then serves as the ‘action phase’ for education acting upon these established plans, ultimately leading to the ‘post-action pursuing phase’ of engaging in actual employment in adulthood (Lent et al., 1994; Lent et al., 2018). Accordingly, it is theoretically and practically valid for the current study to investigate how the evolving trajectories of students’ science performance in middle school may lead to their successful graduation with a STEM degree and later engagement in STEM professions. This pathway is hypothesized to result from the collective impact of the aforementioned education-related socialization agents—at parental, peer, school, and individual student levels—working concomitantly. This is important and relevant as students in the formative years of middle school are required to study core science subjects, such as biology, physics, chemistry, and environmental science, which are all prerequisites for their later, more advanced STEM development in adulthood (Ocal et al., 2021; Song & Wang, 2021).

2.1. The Synergistic Influence of Socialization Agents on Science and STEM Development

Students’ educational journeys are not undertaken in isolation but are jointly embedded within a rich social ecology that significantly influences their development. This study focuses on four key socialization agents in seventh grade that collectively create a pro-learning context: parental educational encouragement, peer academic support, a constructive school learning environment, and student self-esteem. While prior research has often examined these factors in isolation, their synergistic influence is paramount. SCCT and ecological models suggest that these agents operate concurrently across multiple levels (family, peer, school, and individual) to shape students’ academic self-concept, motivation, and performance (Bottia et al., 2021; Sahin et al., 2018; Woo et al., 2021).

To ensure conceptual clarity, we define and operationalize each agent as follows:

• Parental Educational Encouragement is conceptualized as the active promotion of academic values and future educational aspirations by parents or primary caregivers (Dotterer, 2022; Masud et al., 2015). In this study, it is operationalized specifically as parental college push—the extent to which parents signal the importance of and encourage their child to pursue a college education (Miller & Pearson, 2012). This measure captures a form of academic socialization that is more focused on long-term educational goals rather than day-to-day emotional support.
• Peer Academic Support is distinguished from general friendship or social acceptance. It refers to the perceived level to which a student’s friends value and support academic effort, educational planning, and the goal of attending college (Altermatt, 2019; Hoferichter et al., 2022). This construct taps into the normative academic climate within the peer microsystem, where pro-academic attitudes and behaviors are reinforced.
• Constructive school learning environment encompasses the structural and relational qualities of the school that facilitate learning and scholarly development (Stewart, 2008; Zeng et al., 2022). It is measured by parental perceptions of overall school quality, teacher subject-matter expertise, and the degree to which teachers are perceived as caring about students (M. T. Wang & Eccles, 2013; K. Y. Zhao et al., 2023). This agent represents the formal instructional and emotional support system provided by the school institution.
• Student Self-Esteem is conceptualized in this study as a foundational, global personal resource—a relatively stable trait reflecting an individual’s overall sense of self-worth and satisfaction (Booth & Gerard, 2011; Cvencek et al., 2018). While it can be influenced by socialization experiences, we position it here as a key individual-level asset that provides the psychological capital necessary for students to engage with academic challenges and plan for the future, which in turn can be bolstered by positive feedback from parents, peers, and teachers.

Explicitly, parents, peers, school, and students’ own individual resources (e.g., self-esteem) collectively form a socialization context to impact students’ science and STEM development (Bottia et al., 2021; Sahin et al., 2018; Woo et al., 2021; Yeung & Xia, 2023). It is hence research-worthy to include these education-related socialization agents together to scrutinize their joint effects on students’ science performance in middle school and their later STEM development in adulthood. This is consonant with the SCCT and Life Course Theories that educational and career outcomes of students are an integrated continuum that is concomitantly shaped by students’ environmental and individual factors (Gecas, 2003; Lent & Brown, 2013; Lent et al., 1994). Although researchers have examined some of these socialization agents individually in relation to students’ science and STEM development (Booth & Gerard, 2011; Bottia et al., 2021; Mitsopoulou & Pavlatou, 2021; Sahin et al., 2018), to the authors’ knowledge, no research has attempted to simultaneously integrate them in a single investigation to examine their collective impact on students’ science performance in middle school and subsequent STEM development in adulthood.

Investigating the collective impact of these socialization agents is pivotal. The learning contexts created by family, school, peer networks, and students’ personality traits are inseparable, and all influence educational achievement and career direction (Bottia et al., 2021; Y. Chen et al., 2022; Luo et al., 2022; Woo et al., 2021). In this study, we build upon existing, albeit indirect, empirical evidence to examine the interconnected longitudinal relationships linking these socialization agents to students’ science performance during middle school and later STEM development in adulthood. In their longitudinal study, Yeung and Xia (2023) found that parental education-related support and the individual factor of students’ educational expectations at grade 7 and students’ school commitment at grade 8 significantly predicted better academic performance of students in grade 9, including science scores, and later educational achievement in adulthood, including graduating with a STEM major. Moreover, Stewart (2008) investigated a representative sample of 10th graders in the United States and reported that positive parental and peer supports, constructive school climate, and students’ individual efforts significantly contributed to the students’ higher grades in math, English, history, and science. Pertinent to the STEM development of students in adulthood, Iroaganachi et al. (2021) found that parental and teachers’ educational socialization and academic role models from significant others significantly predicted students’ higher STEM path intentions in a sample of 361 junior high school girls. In addition, Quintana and Saatcioglu (2022) recently employed representative data from the High School Longitudinal Study of 2009 (HSLS:09) and corroborated that students’ grade-9 science identity measured by identifying themselves as ‘science kind of person’ was significantly related to their majoring in STEM degree study and a future career plan for the STEM field after college graduation. In that study, the authors included covariates such as students’ mind-set beliefs, educational aspirations, parental academic encouragement, teachers’ educational support, and peers’ math and science performance to demonstrate their relevance, although their specific significant effects on later STEM outcomes were omitted for simplicity. In a recent review of 50 studies, Bottia et al. (2021) examined the social and individual factors affecting STEM participation among racially minoritized college students. They identified poorer family and social capital—manifested as a lack of positive academic peer interactions, weaker secondary school preparation, and reduced psychological competence—as key factors associated with lower STEM engagement. Collectively, this research underscores that a supportive socialization context, encompassing parents, peers, schools, and the individual, is crucial for developing the science proficiency in middle school that forms the foundation for long-term STEM achievement.

Using a collective approach to examine how these socialization agents inextricably interweave a pro-learning context is both pertinent and important for understanding students’ educational path and STEM development. Specifically, parental encouragement provides essential emotional support and signals the value of education, directly influencing adolescents’ academic expectations and performance (Boonk et al., 2018; Tey et al., 2020). Peers become increasingly influential during early adolescence, serving as sources of academic motivation, collaboration, and normative standards; peer support has been specifically linked to better science performance and stronger STEM intentions (Gong et al., 2022; Tey et al., 2020). The school environment, characterized by high-quality instruction and caring teachers, provides the formal structure and encouragement necessary for engaging with challenging scientific concepts and skills (Canales & Maldonado, 2023; Ketscher et al., 2025; LeBeau et al., 2012). Finally, positive self-esteem—a global sense of self-worth formed in early life—provides the foundational psychological capital that enables students to confront academic challenges, plan for the future, and develop more specific competencies like academic self-efficacy (Becker et al., 2021; Magnusson & Nermo, 2018; Scherrer & Preckel, 2019). Therefore, this study argues that these four socialization agents constitute an integral developmental context. Their collective impact in seventh grade is hypothesized to initiate a positive cascade, fostering the science learning trajectories throughout middle school that are essential for distal adult outcomes such as obtaining a STEM degree and securing STEM employment.

2.2. Science Learning Trajectories: Capturing a Dynamic Developmental Process

A pivotal advancement in educational research is the reconceptualization of academic achievement not as a static attribute but as a dynamic and malleable process (Arens et al., 2017; Starr et al., 2022; Yeung & Igarashi, 2025). This is particularly true for science learning in middle school, a period where students are introduced to the core disciplines of biology, physics, chemistry, and environmental science that form the essential groundwork for all later advanced STEM learning (English, 2017; Ocal et al., 2021). To capture this dynamism, this study employs the concept of developmental and growth trajectories, which encapsulate both a student’s initial proficiency level (intercept) at the beginning of middle school and their rate of change (slope) over the subsequent three years. Recent, more advanced longitudinal research has reported that the educational performance of students in secondary school is an evolving and changeable process rather than fixed and stable (Pekrun et al., 2017; Schober et al., 2018), for which the development and changeability of students’ science performance in middle school are also corroborated (Schober et al., 2018; Yeung & Igarashi, 2025). Undeniably, the science performance of students in middle school connotes their establishment of the most foundational and essential basic science knowledge and skills for future STEM development in adulthood (Larkin & Lowrie, 2022; Ozulku & Kloser, 2023). This longitudinal approach offers a more nuanced understanding than single-point assessments. A high initial level of achievement may reflect strong prior preparation, while a positive growth rate indicates successful knowledge acquisition and skill development during middle school itself. Both components are critical. As Mau et al. (1995) aptly noted, preparation for STEM fields must begin early, and mastery of fundamentals must be secured before high school. While some studies have linked high school science scores to STEM major enrolment (Herskovic & Silva, 2022; Kohen & Nitzan, 2022), they often treat science performance as a fixed predictor, neglecting its developmental and evolving nature during the formative middle school years.

Furthermore, enrolment in a STEM major in college does not necessarily secure students’ subsequent successful acquisition of a STEM degree (Kohen & Nitzan, 2022; Miller & Pearson, 2012), in which the latter is more directly relevant to students’ engagement in STEM careers. Empirically, limited empirical research has pointed out that obtaining a STEM degree in college is an important precursor of students’ later STEM employment (Delaney & Devereux, 2022; Wright et al., 2017). This is congruent with the SCCT and Life Course Theories positing that STEM development of students is a continuum (Gecas, 2003; Lent & Brown, 2013; Lent et al., 1994), in which early life stages and pre-action planning and action phases of students, e.g., students’ development and growth of science performance in middle school and gradation with a STEM degree in college, may jointly shape students’ later engagement in STEM professions in adulthood (Ahmed, 2018; Wright et al., 2017). Thus, this study posits that the developmental and growth trajectories of science learning in middle school are a key mechanism—a latent evolving process—through which the early socialization context influences the likelihood of overcoming these bottlenecks. These trajectories represent not just an individual’s standing relative to peers but also their intrapersonal growth across a critical educational period (Newsom, 2024). Nevertheless, existing research has seldom treated the effects of students’ science performance during middle school as evolving trajectories and examined their impacts on students’ later STEM development in adulthood (Kohen & Nitzan, 2022; Ocal et al., 2021; Yeung & Xia, 2023). Although some short-term longitudinal studies have considered the developmental and changing nature of students’ science performance in early years, these studies only employed a cross-lagged design to examine the development and changes in students’ science performance by a rank-ordered fashion (Arens et al., 2017; Marsh, 2022; Pekrun et al., 2017), which are unable to scrutinize the complex and dynamic interconnections between students’ science performance in middle school and their later STEM development in adulthood.

To capture the dynamic nature of science learning, this study conceptualizes performance not as a static point but as an evolving trajectory, modeled latently from repeated measures across grades 7 to 9. This trajectory encapsulates two dimensions: the student’s comparative standing among peers and their unique, intrapersonal rate of growth (Grimm et al., 2017; Ocal et al., 2021). This progression is theorized to culminate in a pivotal distal outcome: the successful attainment of a STEM degree, defined here as a four-year baccalaureate in science, technology, engineering, mathematics, or medicine, which serves as the formal gateway to advanced STEM career pathways. In addition, students’ engagement in STEM professions in adulthood means their employment in the fields of science, medical science, health and life science, mathematics, and engineering after graduation or leaving school as compared to their classmates otherwise. This dichotomous classification of the two STEM outcomes mentioned above has been used in existing research (Ahmed, 2018; Miller & Pearson, 2012; Wright et al., 2017). In conclusion, by integrating the collective influence of seventh-grade socialization agents with a dynamic modeling of science learning trajectories, this study provides a comprehensive test of the theoretical proposition that STEM development is a continuum and developmental course. It traces the pathways from the proximal social environment in early adolescence, through the academic development in middle school, to the ultimate realization of STEM career success in adulthood, thereby addressing a significant gap in the literature. To clarify the integration of these theoretical perspectives and the derived hypotheses, Figure 1 provides a schematic diagram (see Section 3 below) illustrating how SCCT and Life Course Theories inform the selection of the four key socialization agents and their proposed influence on science trajectories and later distal STEM outcomes.

Figure 1. Graphic display of the hypothesized longitudinal relationships between grade-7 socialization agents, students’ developmental and growth trajectories of science performance in middle school, and STEM development in adulthood. Note: The H1a to H3 hypotheses represent direct effects, and the H4a to H5d hypotheses in parentheses are the indirect effects. ScPerf is students’ science performance in middle school. The measures of students’ integrated science performance scores from grade 7 to grade 9 are treated as the multiple indicators to project the developmental and growth trajectories of students’ science performance during middle school, in which they are all fixed to 1 and loaded on the intercept factor to represent the developmental trajectory of students’ science performance in middle school, and are set as [0, 1, 2] for the equal time intervals and loaded on the slope factor to connote the growth trajectory of students’ science performance across middle school. More detailed explanations of the modeling procedures are discussed in the section ‘Modeling Procedures’.

3. The Present Study

Guided by the integrative frameworks of Social Cognitive Career Theory (SCCT) and Life Course Theory, and addressing the identified gaps in the literature, the present study aims to chart the longitudinal pathway from early adolescent socialization to adult STEM career success. Specifically, this investigation examines how a constellation of seventh-grade socialization agents—parental educational encouragement, peer academic support, a constructive school learning environment, and student self-esteem—synergistically shape the evolving trajectories of science performance throughout middle school. Furthermore, it tests whether these dynamic trajectories, in turn, predict the attainment of a STEM college degree and ultimately mediate engagement in a STEM profession in adulthood.

To this end, we posit a sequential mediation model that traces the influence of the grade-7 socialization context through two key academic milestones. First, we hypothesize that the initial level (intercept) and growth rate (slope) of science achievement across middle school will mediate the relationship between the socialization agents and successfully graduating with a STEM degree. Second, we propose that obtaining a STEM degree, preceded by middle school science trajectories, will further mediate the link between the early socialization context and adult STEM career engagement. This model captures the continuum and progression of STEM development, from the “pre-action planning phase” in middle school to the “action phase” of degree completion and the “post-action pursuing phase” of career entry. Guided by the integrative SCCT and Life Course Theory framework outlined above, which posits a sequential pathway from early socialization through academic development to career attainment, we formulated a set of five hypotheses. These hypotheses are designed to empirically test the proposed structural growth model, examining the direct, mediation, and sequential relationships between the grade-7 socialization context, middle school science learning trajectories, and adult STEM outcomes. Based on this theoretical rationale, the following hypotheses are advanced:

H1. 

Grade-7 socialization agents of parental educational encouragement, academic support of peers, constructive school learning environment, and students’ positive self-esteem will positively predict students’ developmental and growth trajectories of science performance in middle school (H1a and H1b), successful graduation with a STEM degree in college (H1c), and later engagement in STEM professions in adulthood (H1d).

H2. 

The developmental and growth trajectories of students’ science performance in middle school will positively predict students’ successful graduation with a STEM degree in college (H2a and H2b) and later engagement in STEM professions in adulthood (H2c and H2d), even accounting for the effects of grade-7 socialization agents.

H3. 

Students’ successful graduation with a STEM degree in college will positively predict students’ later engagement in STEM professions in adulthood, even accounting for the effects of grade-7 socialization agents and students’ developmental and growth trajectories of science performance in middle school.

H4. 

The developmental and growth trajectories of students’ science performance in middle school will mediate the effects of grade-7 socialization agents of parental educational encouragement (H4a), academic support of peers (H4b), constructive school learning environment (H4c), and students’ positive self-esteem (H4d) on students’ successful graduation with a STEM degree in college.

H5. 

The developmental and growth trajectories of students’ science performance in middle school and their successful graduation with a STEM degree in college will mediate the effects of grade-7 socialization agents of parental educational encouragement (H5a), academic support of peers (H5b), constructive school learning environment (H5c), and students’ positive self-esteem (H5d) on students’ later engagement in STEM professions in adulthood.

To ensure the robustness of these analyses, we included key demographic covariates—student gender, family composition, and ethnicity—in all modeling procedures to control for potential confounding effects. Existing research reports that female students, compared to their male counterparts, have higher educational motivation and academic performance, although lower rates of STEM intention and enrolment are found (Delaney & Devereux, 2022; Friedman-Sokuler & Justman, 2016; Wright et al., 2017). For family composition, students living with a biological father and mother in two-parent families compared to their counterparts in other family structures demonstrate better academic and career achievement (Heard, 2007; Sun & Li, 2011), including STEM achievement. For ethnicity, the current study classified student participants into five ethnic groups, including Whites, African Americans, Asians, Hispanics, and Native Americans, with Asian students serving as the reference due to their academic and STEM outperformance (Sahin et al., 2018; Zhang, 2022). To avoid collinearity in modeling, student age is not included as a covariate due to the fact that students’ science performance is treated as a time-varying variable aligned with their school years to longitudinally predict distal STEM outcomes in adulthood (J. Wang & Wang, 2019). Figure 1 shown above provides a visual representation of these hypothesized interconnected longitudinal relationships. By testing this comprehensive growth model, the current study can provide a nuanced understanding of how a supportive middle school ecology fosters science learning and channels students toward long-term STEM success.

4. Methods

4.1. Data Source and Sample

This study utilized data from the Longitudinal Study of American Youth (LSAY), a nationally representative study designed to track the educational and occupational development of U.S. public school students, with a specific emphasis on pathways into science and engineering (Miller, 2014). Initiated in 1987, LSAY was designed to trace the educational and occupational pathways of students, with a particular emphasis on scientific development and long-term STEM attainment. The study initially recruited two nationally representative cohorts: one consisting of 2829 tenth-grade students and another comprising 3116 seventh-grade students. These participants were surveyed annually for seven years, from 1987 to 1994, resulting in longitudinal data that covered three years of high school and four years of post-secondary transition for the first cohort, and six years of secondary education plus one-year post-high school for the second. In 2006, the National Science Foundation funded an additional follow-up phase, enabling five further waves of data collection between 2007 and 2011. This extended timeline facilitated the remarkable retention of approximately 95% of the original combined sample (N = 5945) by 2007, when participants were between 33 and 37 years of age. To ensure national representativeness LSAY employed a stratified sampling design, selecting public middle and high schools from 12 strata defined by geographic region and community type.

The analysis draws on the cohort of 3116 students who were first surveyed in the seventh grade in 1987 (Cohort 2). This cohort was followed annually through middle and high school until 1993 and participated in additional follow-up surveys in adulthood between 2007 and 2011. The present analysis leverages this extensive longitudinal design, covering a period of over 20 years from early adolescence to mid-adulthood (age 33–37 in 2007 and age 37–41 in 2011).

4.2. Measures

All constructs were measured using established items from the LSAY protocol (Miller, 2014). All data were collected and matched at the individual participant level across all waves of the study, creating a longitudinal record for each of the 3116 students.

Distal Outcomes in Adulthood:

-

Engagement in STEM Professions was assessed in 2008 with a dichotomous measure (0 = no, 1 = yes) indicating whether the participant was employed in a STEM field, as classified by LSAY personnel. This classification encompassed careers in science, engineering, mathematics, medical science, and health and life sciences (Ahmed, 2018). This outcome aligns with the post-action pursuing phase in SCCT and the realization of long-term trajectories as per Life Course Theory.

-

Graduation with a STEM Degree was measured in 2007, indicating whether the participant had successfully obtained a four-year baccalaureate degree majoring in science, technology, engineering, mathematics, or medicine (0 = no baccalaureate or non-STEM major, 1 = STEM major) (Kang et al., 2021; Wright et al., 2017). This milestone corresponds to the action phase in the SCCT framework.

Time-Varying Measure:

-

Science performance was assessed annually from grades 7 to 9 (1987–1989) using integrated science test scores provided by LSAY. These scores, scaled using Item Response Theory (IRT) to adjust for difficulty and reliability (DeMars, 2010), measured proficiency in core scientific disciplines (biology, physical science, chemistry, environmental science) essential for future STEM development. Scores ranged from 26 to 88 in grade 7 to 27–91 in grade 9, with higher scores indicating better performance. These scores represent the evolving performance attainment and development of academic self-efficacy central to SCCT, captured as a dynamic trajectory.

Grade-7 Socialization Agents (Time-Invariant Covariates):

-

Parental educational encouragement was measured in grade 7 in the year 1987 with an item of “parent college push in grade 7” that was rated by the main caregiver parent with a 6-point scale ranging from 0 = low parent college push to 5 = high parent college push, in which higher scores indicate more parental educational expectations and encouragement for students’ higher education. This measure has been used in research to tap into parents’ educational support and expectations for their children to attain a college education (Miller & Pearson, 2012). This LSAY item (parent college push) directly operationalizes the environmental support and contextual affordances emphasized in SCCT and Life Course Theories.

-

Peer academic support was measured in grade 7 in the year 1987 with an item of “peer academic push in grade 7” that was rated by the student participants with a 4-point scale ranging from 0 = low academic push to 3 = high academic push to indicate whether friends of the student participant supported the importance of academic performance, educational plan, and college education. Recent research has used similar measures to rate academic support from peers (Altermatt, 2019; Hoferichter et al., 2022). This item (peer academic push in grade 7) serves as a proxy for the peer influence and normative environment within the student’s social context, as outlined in ecological models and SCCT.

-

Constructive school learning environment was measured in grade 7 in the year 1987 with three items that include (1) “Overall positive quality of the school”, (2) “Teachers know their subjects”, and (3) “Teachers care about students”, which were rated by the main caregiver parent on a 5-point scale from 1= strongly agree to 5 = strongly disagree. The three items were reverse-coded and averaged to recreate a composite score, with higher scores representing a more positive and pro-learning school environment. Recent research has adopted a similar approach to measure positive learning environments in schools for students (Zeng et al., 2022; K. Y. Zhao et al., 2023). This composite scale captures the environmental support and learning climate at the school level, a key contextual factor in SCCT and Life Course Theories. Cronbach alpha is reliable, α = 0.766.

-

Positive self-esteem of students was measured in grade 7 in the year 1987 with four items that include (1) “positive attitude toward self”, (2) “I am a person of worth”, (3) “able to do things as well as others”, and (4) “generally satisfied with self”, which were rated by the student participants on a 5-point scale from 1 = strongly agree to 5 = strongly disagree and averaged as a composite score by inverse coding to indicate higher scores representing better self-esteem. Prior research has employed a similar measure to assess the positive self-esteem of adolescent students (Regnerus & Elder, 2003). This scale measures the personal input factor of global self-worth, which provides the foundational psychological capital for the development of domain-specific self-efficacy as per SCCT. Cronbach alpha is reliable, α = 0.706.

Demographic Covariates:

-

Several time-invariant covariates were included to control for potential confounding effects: gender (0 = female, 1 = male), family composition (0 = other structures, 1 = two-biological-parent family), and ethnicity, which was used to classify student participants into five ethnic groups (Whites, African Americans, Hispanics, Asians, and Native Americans), with Asian students serving as the reference (0) and the remaining ethnic groups as comparison (1) to create four dummy ethnic variables.

4.3. Modeling Procedures

To test the hypothesized longitudinal relationships, we employed conditional latent growth curve modeling (LGCM) with time-invariant covariates and distal outcomes (McArdle, 2012; Smid et al., 2020). LGCM is a powerful structural equation modeling technique ideal for capturing intra-individual change over time. It estimates two key latent factors for each individual, which include the intercept (η_0i) representing the initial level of science achievement at the beginning of the growth trajectory (grade 7), and the slope (η_1i) representing the rate of linear change (growth) in science achievement across the three middle school years (grades 7 to 9). The unconditional LGCM for the repeated measures of science performance ($y_{ti}$) for individual i at time t is represented by the following equation:

$$y_{ti}={\alpha }_t+{\lambda }_{0t}{\eta }_{0i}+{\lambda }_{1t}{\eta }_{1i}+{\epsilon }_{ti},$$

where

-

$y_{ti}$ is the observed science performance for individual $i$ at grade $t$.

-

${\alpha }_t$ is the intercept of the science performance at time $t$.

-

${\lambda }_{0t}$ and ${\lambda }_{1t}$ are the factor loadings for the intercept and slope factors of science performance, respectively. To define a linear trajectory, the intercept factor loadings were fixed to 1 for all time points, and the slope factor loadings were fixed to 0, 1, 2 for grades 7, 8, and 9.

-

${\eta }_{0i}$ and ${\eta }_{1i}$ are the latent intercept and slope factors of science performance for individual $i$.

-

${ϵ}_{ti}$ is the time-specific residual for individual $i$ at time $t$, with variance ${\sigma }_e^2$.

To address our research questions, we extended the model to a conditional LGCM model to incorporate the grade-7 socialization agents and demographic variables. The latent intercept and slope factors were regressed on the time-invariant covariates—the grade-7 socialization agents (parental encouragement, peer support, school environment, and students’ self-esteem) and demographic controls (gender, family composition, ethnicity). This is formalized by the equations:

$${\eta }_{0i}={\gamma }_{00}+{\gamma }_{01}{z}_i+{\zeta }_{0i}$$

$${\eta }_{1i}={\gamma }_{10}+{\gamma }_{11}{z}_i+{\zeta }_{1i},$$

where

-

γ₀₀ and γ₁₀ are the mean intercept and mean slope for the developmental and growth trajectories of science performance, respectively.

-

z_i is a vector of the time-invariant covariates for individual i.

-

γ₀₁ and γ₁₁ are vectors of regression coefficients representing the effects of the covariates in terms of the socialization agents and controlled demographic variables on the latent intercept and slope factors.

-

ζ_0i and ζ_1i are the residuals (random effects) of the intercept and slope factors.

Finally, to predict the distal adult outcomes (obtaining a STEM degree and STEM career engagement), these ordered categorical variables (η_Di) were regressed onto the latent growth factors (η_0i, η_1i) and the vector of grade-7 covariates (z_i) in a single, integrated LGCM model:

$${\eta }_{Di}=a_D+{\beta }_1{\eta }_{0i}+{\beta }_2{\eta }_{Ii}+{\beta }_3{z}_i+{\xi }_{Di},$$

where

-

β₁ and β₂ quantify the direct effects of the initial science performance level and its growth rate on the distal outcome, e.g., STEM degree attainment and STEM career engagement.

-

β₃ represents the direct effects of the socialization agents on the distal outcome, controlling for the science trajectories.

-

ξ_Di is the residual variance of the distal outcome.

All modeling procedures were estimated in Mplus 8.11 (Muthen & Muthen, 1998–2017) using a robust weighted least squares estimator (WLSMV) suitable for categorical distal outcomes. To account for the nested structure of the LSAY data (students nested within schools), the <TYPE = COMPLEX> option was used to adjust standard errors and model fit statistics (Heck & Thomas, 2020). Missing data were handled using full information maximum likelihood (FIML) under the assumption of missing at random (MAR), which utilizes all available data for parameter estimation (Lee & Shi, 2021). Recent Monte Carlo simulation studies reported that FIML and multiple imputation (MI) techniques produce similar results in terms of power, bias, and efficiency (Lee & Shi, 2021). Following conventional criteria, acceptable fit was indicated by CFI > 0.90, RMSEA < 0.08, and SRMR < 0.10, while excellent fit was indicated by CFI > 0.95, RMSEA < 0.06, and SRMR < 0.08 (Heck & Thomas, 2020; J. Wang & Wang, 2019).

5. Results

5.1. Descriptive Statistics of the Study Variables

Table 1 presents the descriptive statistics of the study variables. Female and male students shared 48% (n = 1495) and 52% (n = 1621, respectively, and the majority of the students were from two-parent families (87.1%, n = 2715) compared with those of other family structures (12.9%, n = 401). For ethnicity, the majority were Whites (n = 2169, 69.6%), followed by African Americans (n = 504, 16.2%), Hispanics (n = 284, 9.1%), Asians (n = 112, 3.5%), and Native Americans (n = 47, 1.5%). The demographic characteristics presented in this sample are consistent with the original sample of LSAY. Moreover, the mean of parental educational encouragement was 1.796 measured on a 6-point scale (0 = low parent college push to 5 = high parent college push), which suggests that the parents of students did not show a strong academic push for the educational development of their student children. In contrast, the mean of peer academic support was 2.035 based on a 4-point scale (0 = low academic push to 3 = high academic push), indicating that the classmate peers of students were generally more educationally supportive. In addition, the means of constructive school learning environment and students’ positive self-esteem, measured on a 5-point scale (1= strongly agree to 5 = strongly disagree), were 2.333 and 3.922, respectively. This indicates that the supportive learning environment in students’ schools was average, but their positive self-concept was relatively high. In addition, the average integrated scores of students’ science performance show an ascending trend across middle school, which are $\overline{x}$ = 49.854 in grade 7, $\overline{x}$ = 53.282 in grade 8, and $\overline{x}$ = 57.440 in grade 9. However, only a small number of the students successfully graduated with a four-year STEM degree (n = 281, 9%) observed in 2007 and engaged in STEM professions (n = 191, 6.1%) in 2008 afterward.

Table 1. Descriptive statistics of the study variables (N = 3116).

Study Variables		Mean/Frequency	SD/%
1.	Gender
	Female	1495	48%
	Male	1621	52%
2.	Family Composition
	Two-Parent	2715	87.1%
	Otherwise	401	12.9%
3.	Ethnicity
	Non-Hispanic White	2169	69.6%
	American African	504	16.2%
	Hispanic	284	9.1%
	Asian	112	3.6%
	Native	47	1.5%
4.	Grade-7 Parental Educational Encouragement in 1987	1.796	1.064
5.	Grade-7 Peer Academic Support in 1987	2.035	0.998
6.	Grade-7 Constructive School Learning Environment in 1987	2.333	0.613
7.	Grade-7 Students’ Positive Self-Esteem in 1987	3.922	0.663
8.	Integrated Scores of Science Performance from 1987 to 1989
	Grade 7 (1987)	49.854	10.213
	Grade 8 (1988)	53.282	10.814
	Grade 9 (1989)	57.440	10.929
9.	STEM Degree in 2007
	Yes	281	9%
	No	2835	91%
10.	STEM Professions in 2008
	Yes	191	6.1%
	No	2925	93.9%

Note. Demographic covariates of students’ gender, family composition, and ethnicity were collected in grade 7. Parental educational encouragement was measured on a 6-point scale (0 = low parent college push to 5 = high parent college push), peer academic support was measured on a 4-point scale (0 = low academic push to 3 = high academic push), and constructive school learning environment and students’ positive self-esteem were measured on a 5-point scale (1 = strongly agree to 5 = strongly disagree).

5.2. Results of Unconditional and Conditional LGCM Modeling

First, an unconditional LGCM model was conducted to examine the developmental and growth trajectories of students’ science performance in middle school (Model 1). Table 2 shows that the latent intercept and slope factors were significantly loaded on students’ science performance across middle school years, in which the intercept factor loadings of grades 7, 8, and 9 were λ = 0.964, 0.908, and 0.904, p < 0.001, and the slope factor loadings of grades 8 and 9 were λ = 0.292 and 0.581, p < 0.001. The correlation of the science performance intercept and slope factors was r = −0.134, p < 0.001, indicating that students with higher initial level of science performance in middle school exhibited a slower rate of growth in science performance across the middle school years. This unconditional model fits the data well: CFI = 1.000, RMSEA = 0.000, and SRMR = 0.000.

Table 2. Unconditional latent growth curve modeling of students’ science performance in middle school across grades 7, 8, and 9 (Model 1).

Intercept Factor	Factor Loadings	SE	95% CI
Grade-7 Science Performance	0.964 ***	0.013	0.939 to 0.988
Grade-8 Science Performance	0.908 ***	0.011	0.887 to 0.929
Grade-9 Science Performance	0.904 ***	0.016	0.872 to 0.935
Slope Factor	Factor Loadings	SE	95% CI
Grade-7 Science Performance	0.000	0.000	--
Grade-8 Science Performance	0.292 ***	0.012	0.268 to 0.316
Grade-9 Science Performance	0.581 ***	0.026	0.530 to 0.633
Model Parameters	Coefficients	SE	95% CI
Intercept and Slope Covariance	−0.134 ***	1.139	−6.435 to −1.970
Intercept Variance	97.419 ***	4.229	89.130 to 105.709
Slope Variance	10.080 ***	1.044	8.033 to 12.126

Note: For model fit, CFI = 1.000, RMSEA = 0.000, SRMR = 0.000. The factorings loadings and intercept and slope covariance are presented in standardized values. * p < 0.05, ** p < 0.01, *** p < 0.001.

Next, a conditional LGCM model with time-invariant covariates was conducted to test the effects of grade-7 socialization agents in terms of parental educational encouragement, peer academic support, constructive school learning environment, and students’ positive self-esteem on the developmental and growth trajectories of students’ science performance in middle school while adjusting for the demographic variables of students’ gender, family composition, and ethnicity (Model 2). In this model, we set the relationship between the science performance intercept and slope factors as directional rather than correlational by regressing the latter on the former to test whether the growth of students’ science performance is a function of their initial development of science performance in middle school. An excellent model fit was obtained for Model 2: CFI = 1.000, RMSEA = 0.000, and SRMR = 0.004.

Figure 2 shows the results of Model 2 that grade-7 socialization agents significantly predicted students’ higher development of science performance in middle school, in which parental educational encouragement had the strongest positive effect, β = 0.230, p < 0.001, followed by peer academic support, β = 0.118, p < 0.001, students’ positive self-esteem, β = 0.059, p < 0.01, and constructive school learning environment, β = 0.053, p < 0.01. Nevertheless, only the socialization agents of parental educational encouragement (β = 0.067, p < 0.01) and constructive school learning environment (β = 0.041, p < 0.05) significantly predicted students’ positive growth in science performance across the middle school years. In addition, the growth of students’ science performance was significantly found as a function of students’ initial development of science performance in middle school (β = −0.146, p < 0.001), indicating that students with higher initial development of science performance displayed a lower rate of growth in science performance later in middle school years. For the effects of students’ demographic covariates, male students and students of two-parent family status significantly had a higher developmental trajectory of science performance in middle school, β = 0.196 and 0.039, p < 0.001 and 0.05 (see Table S1 in the Supplementary Materials). Compared to their Asian classmates, Hispanic, African American, and Native American students had a significantly lower developmental trajectory of science performance in middle school: β = −0.176, −0.247, −0.060, p < 0.001, 0.001, and 0.05. However, female students, in contrast to their male peers, had significantly higher growth in science performance across middle school years, β = −0.057, p < 0.05.

Figure 2. Standardized effects of conditional latent growth curve modeling predicting developmental and growth trajectories of students’ science performance by grade-7 socialization agents (Model 2). Note: ScPerf is students’ science performance in middle school. Students’ demographic covariates in terms of gender, family composition, and ethnicity were incorporated in the modeling procedures. For simplicity, the effects of students’ gender, family composition, and ethnicity on the intercept and slope factors of students’ science performance in middle school were skipped in the figure, and insignificant regression effects were presented in dotted arrows. The model fit is CFI = 1.000, RMSEA = 0.000, SRMR = 0.004. * p < 0.05, ** p < 0.01, *** p < 0.001.

Furthermore, a conditional integrated LGCM model with time-invariant covariates and the two distal STEM outcomes was conducted to test the effects of grade-7 socialization agents and students’ developmental and growth trajectories of science performance in middle school on students’ successful graduation with a STEM degree in college and later engagement in STEM professions (Model 3), while simultaneously adjusting for students’ demographic covariates of gender, family composition, and ethnicity. Model 3 fits the data well: CFI = 1.000, RMSEA = 0.000, and SRMR = 0.003. Figure 3 shows the results of Model 3 that the socialization agent of parental educational encouragement measured in grade 7 not only significantly predicted students’ better developmental and growth trajectories of science performance during middle school years, β = 0.230 and 0.067, p < 0.001 and 0.01, but also directly contributed to students’ successful completion of a STEM degree in college, β = 0.070, p < 0.001, indicating that a unit increase in parental educational encouragement resulted in the higher odds of students’ successful graduation with a college STEM degree by 7.25%. In addition, peer academic support had significant direct effects on the better development of students’ science performance in middle school (β = 0.118, p < 0.001) and their successful graduation with a STEM degree in college (β = 0.037, p < 0.05), for which the latter connotes that a unit increase in peer academic support contributed to the higher odds of students’ successful graduation with a college STEM degree by 3.76%.

Figure 3. Standardized effects of conditional latent growth curve modeling predicting distal STEM out-comes by developmental and growth trajectories of students’ science performance and grade-7 socialization agents (Model 3). Note: ScPerf is students’ science performance in middle school. Students’ demographic covariates in terms of gender, family composition, and ethnicity were incorporated in the modeling procedures. For simplicity, the effects of students’ gender, family composition, and ethnicity on the intercept and slope factors of students’ science performance in middle school and students’ distal STEM outcomes were skipped in the figure. Insignificant regression effects were presented in dotted arrows. The model fit is CFI = 1.000, RMSEA = 0.000, SRMR = 0.003. * p < 0.05, ** p < 0.01, *** p < 0.001.

Moreover, constructive school learning environment significantly predicted students’ better developmental and growth trajectories of science performance during middle school years, β = 0.053 and 0.041, p < 0.01 and 0.05. Also, students’ positive self-esteem had significant effects on their better development of science performance in middle school (β = 0.059, p < 0.01) and later engagement in STEM professions in mid-adulthood (β = 0.033, p < 0.05), for which the latter indicates that a unit increase in students’ self-esteem led to the higher odds of engagement in STEM professions by 3.35%. In addition, the slope factor of students’ science performance was significantly and negatively predicted by its intercept factor, β = −0.145, p < 0.001, which connotes that students with higher initial development of science performance in middle school demonstrated a lower rate of growth in science performance in later middle school years.

Furthermore and importantly, both the intercept and slope factors of students’ science performance in middle school significantly and positively predicted students’ successful graduation with a STEM degree in college, β = 0.208 and 0.067, p < 0.001 and 0.01, indicating that a unit increase in the developmental and growth trajectories of students’ science performance during middle school years contributed to their higher odds of acquisition of a STEM degree in college by 22.51% and 6.93%, respectively. In addition, the intercept factor of students’ science performance in middle school and their successful graduation with a STEM degree in college significantly and positively predicted students’ later engagement in STEM professions in adulthood, β = 0.059 and 0.608, p < 0.01, meaning that a unit increase in students’ better development of science performance in middle school and their successful acquisition of a STEM degree in college resulted in the higher odds of students’ later engagement in the STEM professions by 6.07% and 83.67%, respectively.

On the other hand, the effects of students’ demographic covariates—gender, family composition, and ethnicity—on the developmental and growth trajectories of students’ science performance in middle school were similar to those of Model 2 (see Table S1 in the Supplementary Materials). A notable finding was that while male students were significantly more likely to graduate with a STEM degree in college (β = 0.049, p < 0.05), indicating that their odds were 5.02% higher than their female classmates, a significant negative effect of gender was found on STEM career engagement (β = −0.032, p < 0.05), which indicates that female students were more likely to engage in STEM professions than male students. Additionally, compared to their Asian peers, Hispanic, African American, and White students were all less likely to graduate with a STEM degree (β = −0.102, −0.129, and −0.119, p < 0.01, 0.01, and 0.05, respectively). These coefficients indicate decreased odds of obtaining a STEM degree in college by 9.69%, 12.10%, and 11.21% for Hispanic, African American, and White students, respectively.

5.3. Latent Serial Mediation Effects of Students’ Developmental and Growth Trajectories of Science Performance and Their Graduation with a STEM Degree

For testing the latent serial mediation effects of students’ developmental and growth trajectories of science performance in middle school and/or their graduation with a STEM degree on the longitudinal relationships between the grade-7 socialization agents and the two distal STEM outcomes in adulthood, total indirect effects were reported here. Total indirect effects can more accurately present the aggregated effect sizes of all related mediators and mediational paths simultaneously. Table 3 shows that the total indirect effects of parental educational encouragement, peer academic support, constructive school learning environment, and positive self-esteem of students on students’ successful graduation with a STEM degree in college via the mediations of intercept and slope factors of students’ science performance were β_TotInd = 0.050, 0.023, 0.013, and 0.013, p < 0.001, 0.001, 0.01, and 0.01, respectively (see Table S2 in the Supplementary Materials for individual indirect effects). This means that the developmental and growth trajectories of students’ science performance in middle school significantly and jointly mediated the impacts of grade-7 socialization agents on students’ successful graduation with a STEM degree in college.

Table 3. Total indirect effects of socialization agents on students’ successful graduation with a STEM degree in college and engagement in STEM professions in adulthood.

No.	Predictor	Mediators/Mediational Paths a	STEM Outcome	βTotInd	Z-Value
A1	Parental Educational Encouragement	(i) Intercept Factor of Students’ Science performance; (ii) Slope Factor of Students’ Science performance; (iii) Intercept Factor of Students’ Science performance → Slope Factor of Students’ Science performance	Successful Graduation with a STEM Degree	0.050	7.708 ***
A2	Peers Academic Support		Successful Graduation with a STEM Degree	0.023	3.950 ***
A3	Supportive School Learning Environment		Successful Graduation with a STEM Degree	0.013	2.888 **
A4	Positive Self-Esteem of Student		Successful Graduation with a STEM Degree	0.013	3.453 **
B1	Parental Educational Encouragement	(i) Intercept Factor of Students’ Science performance; (ii) Slope Factor of Students’ Science performance; (iii) Successful Graduation with a STEM Degree; (iv) Intercept Factor of Students’ Science performance → Slope Factor of Students’ Science performance; (v) Intercept Factor of Students’ Science performance → Successful Graduation with a STEM Degree; (vi) Slope Factor of Students’ Science performance → Successful Graduation with a STEM Degree; (vii) Intercept Factor of Students’ Science performance → Slope Factor of Students’ Science performance → Successful Graduation with a STEM Degree	Engagement in STEM Professions	0.087	7.323 ***
B2	Peer Academic Support		Engagement in STEM Professions	0.043	3.391 **
B3	Constructive School Learning Environment		Engagement in STEM Professions	0.031	2.625 **
B4	Positive Self-Esteem of Students		Engagement in STEM Professions	0.024	2.004 *

Note. Total indirect effects are the aggregated effect size of all related mediators and mediational paths from the predictor to the STEM outcome variable. All effects are presented in standardized values. * p < 0.05, ** p < 0.01, *** p < 0.001. ^a The hypothesis is completely supported if the effects of its sub-hypotheses or the hypothesis itself are all significantly supported; and the hypothesis is largely supported if more than half of the effects of its sub-hypotheses are significantly supported.

In addition, the total indirect effects of parental educational encouragement, peer academic support, constructive school learning environment, and positive self-esteem of students on students’ engagement in STEM professions via the mediations of the intercept and slope factors of students’ science performance and their successful graduation with a STEM degree were β_TotInd = 0.087, 0.043, 0.031, 0.024, p < 0.001, 0.01, 0.01, and 0.05, respectively (individual indirect effects are reported in Table S3 in the Supplementary Materials). The results confirmed that the developmental and growth trajectories of students’ science performance in middle school and their completion of a STEM degree in college significantly and jointly mediated the relationships between the grade-7 socialization agents and students’ later engagement in STEM professions in adulthood sequentially.

5.4. Summary of Supportiveness of the Study Hypotheses and Sensitivity Analysis

Table 4 summarizes the supportiveness of the five hypothesis sets in the current study. Based on the findings of Model 3, Hypothesis 1 is largely supported, for which the sub-hypothesis H1a is completely supported and the sub-hypotheses H1b, 1c, and 1d are partially supported. Moreover, Hypothesis 2 is also largely supported, in which the sub-hypotheses H2a, 2b, and 2c are all supported, although the sub-hypothesis H2d is not supported. Moreover, Hypothesis 3 is completely supported. In addition, of the results in the latent serial mediation effects found, Hypothesis 4 is completely supported, for which the sub-hypotheses H4a, H4b, H4c, and H4d are all supported. Furthermore, Hypothesis 5 is completely supported, for which the sub-hypotheses H5a, H5b, H5c, and H5d are all tenable.

Table 4. Summary of supportiveness of the five hypotheses set in the study.

Hypothesis	Sub-Hypothesis	Supportiveness	Explanation of Vindications a
Hypothesis 1 (H1)	H1a	✓	Among 16 direct effects being tested (the four socialization agents on students’ development and growth of science performance, graduation with a STEM degree, and engagement in STEM professions), 9 are correctly supported, and 7 are not supported (refer to Model 3 in Figure 2). Hypothesis 1 is largely supported.
	H1b	⍻
	H1c	⍻
	H1d	⍻
Hypothesis 2 (H2)	H2a	✓	Among 4 direct effects being tested (students’ development and growth of science performance on their graduation with a STEM degree and engagement in STEM professions), 3 are correctly supported, and 1 is not supported (refer to Model 3 in Figure 2). Hypothesis 2 is largely supported.
	H2b	✓
	H2c	✓
	H2d	✕
Hypothesis 3 (H3)	--	✓	Hypothesis 3 is completely supported (refer to Model 3 in Figure 2).
Hypothesis 4 (H4)	H4a	✓	The two mediated effects of students’ development and growth of science performance on the relationships between socialization agents and students’ graduation with a STEM degree are tested, they are all correctly supported (refer to Table 3). Hypothesis 4 is completely supported.
	H4b	✓
	H4c	✓
	H4d	✓
Hypothesis 5 (H5)	H5a	✓	The three mediated effects of students’ development and growth of science performance and graduations with a STEM degree on the relationships between socialization agents and students’ engagement in STEM professions are tested, they are all correctly supported (refer to Table 3). Hypothesis 5 is completely supported.
	H5b	✓
	H5c	✓
	H5d	✓

Note. ✓ = completely supported; ⍻ = partially supported; ✕ = not supported. ^a The hypothesis is completely supported if the effects of its sub-hypotheses or the hypothesis itself are all significantly supported; and the hypothesis is largely supported if more than half of the effects of its sub-hypotheses are significantly supported.

To further confirm the stability and validity of the study relationships found in the current study, sensitivity analysis was conducted by replacing the two distal STEM outcomes—graduation with a STEM degree and engagement in STEM professions—with two other STEM-related variables: graduation majors and career categories measured in 2007 and 20081, respectively. For sensitivity analysis, we recoded the variable of graduation majors to three categories that are 0 = students with no undergraduate degree (n = 1164), 1 = students of non-STEM degree (n = 641), and 2 = students of STEM degree (n = 281), and then combined students of non-STEM or STEM degrees to the same category (1) and students with no degree were set as the reference (0), in which the variable was named “Graduation with a (Non-)STEM Degree” to replace the STEM outcome of “Graduation with a STEM Degree” (see Table S4 in the Supplementary Materials for details). The rationale is that STEM development is knowledge-based and progressive (Ekmekci et al., 2019; T. Zhao & Perez-Felkner, 2022), in which students’ science education in middle school constitutes a pivotal subject area of common education in secondary school that can profoundly affect students’ later educational achievement in higher education, including choosing and studying STEM and non-STEM majors. Hence, students with better science performance in middle school and more supportive education-related socialization agents in grade 7 compared to their counterparts otherwise would have a better chance of attaining a STEM or non-STEM degree in college depending on the individual students’ academic interests when selecting majors. In addition, the career categories variable was recoded to replace the distal STEM outcome of “Engagement in STEM Professions”. A broader definition was adopted, which now includes practitioners in science, engineering, health, social sciences, and STEM-support or health-support occupations, who are counted as engaged in STEM professions. This new variable was named “Engagement in STEM Professions (Broad)”, where students engaged in broadly defined STEM careers in mid-adulthood were coded 1 (n = 204), while all other students were coded 0 (n = 2912).

For sensitivity analysis, we reran Model 3 by inputting these two newly coded STEM variables to replace the two original STEM outcomes (termed Model S3 hereafter). If the model fit of Model S3 does not significantly differ with the original Model 3, e.g., ∆CFI ≤ 0.010, ∆RMSEA ≤ 0.015, and ∆SRMR ≤ 0.030, we can confidently accept that the study relationships hypothesized in the current study are held and stable in Model 3 (F. F. Chen, 2007; Marsh et al., 2004). Moreover, the significant effects of Model S3 are expected to be similar to Model 3 but different in magnitude. This is because pooling students of STEM and non-STEM degrees in the same category would augment the effects of students’ development and growth of science performance in middle school and grade-7 education-related socialization agents on students’ successful graduation with a college degree (STEM and non-STEM majors) compared to only examining the effects on students’ graduation with a STEM degree. The rationale is that students with better science performance in middle school may have more jurisdiction to choose a STEM or non-STEM major, in which the stronger effects of students’ developmental and growth trajectories of science performance on students’ acquisition of a STEM or non-STEM degree are expected. In addition, the significant effect of Graduation with a (Non-)STEM degree on students’ Engagement in STEM Professions (Broad) would shrink when pooling students of STEM and non-STEM degree in the same category. This is because students of a non-STEM degree in the same category aligning with students of a STEM degree would still be more likely to opt for non-STEM occupations by self-selection after graduation, hence reducing the variance of engagement in STEM professions broadly defined. However, stronger, and more significant effects of students’ developmental and growth trajectories of science performance in middle school on students’ later engagement in STEM professions are expected due to better science performance in middle school signaling as outperformed core science qualifications for students to choose a STEM or non-STEM major and selecting to work in the STEM field or not after graduation.

Resultantly, sensitivity analysis supported the above-postulated assumptions, in which the model fit of Model S3 is comparable to Model 3: CFI = 1.000, RMSEA = 0.000, and SRMR = 0.004 (see Figure S1 in the Supplementary Materials), and the changes in model fit indexes between Model S3 and Model 3 were: ∆CFI = 0.000, ∆RMSEA = 0.000, and ∆SRMR = 0.001, which indicates no evidence of significant difference between the two models. This supports the stability and validity of the study relationships found in Model 3 even when introducing uncertainty with the replacement of two newly coded STEM variables/outcomes. In addition, the significant and insignificant effects of Model S3 and Model 3 are largely similar, except their difference in magnitude. As expected, Graduation with a (Non-)STEM degree had a smaller significant effect on students’ Engagement in STEM Professions (Broad) in Model S3 compared to Model 3, β = 0.227 vs. 0.608, p < 0.001. In addition, students’ developmental and growth trajectories of science performance in middle school had stronger significant effects on Graduation with a (Non-)STEM degree in Model S3 compared to Model 3, β = 0.323 vs. 0.208, p < 0.001, and β = 0.086 vs. 0.067, p < 0.001 and 0.01. Furthermore, education-related socialization agents were more significantly and strongly predictive of Graduation with a (Non-)STEM degree in Model S3 compared to those in Model 3, in which the effects of constructive school learning environment and students’ self-esteem became significant in Model S3. The similarity of model fit and significance of the effects between Model S3 and Model 3 attest to the stability and validity of the hypothesized interconnections of education-related socialization agents, students’ developmental and growth trajectories of science performance in middle school, and their distal STEM outcomes in adulthood proposed and examined in the current study. Moreover, the latent serial mediation effects of Model S3 are comparable to those of Model 3. All were significant, although they differed slightly in magnitude (see Table S5 in the Supplementary Materials). The full Mplus code and initial result outputs of the current study generated from Mplus programming are available in the Open Science Framework at https://osf.io/7kzv3/overview.

6. Discussion

The present study represents a significant empirical advancement in the educational and developmental sciences by charting a comprehensive, longitudinal pathway from the proximal socialization ecology of early adolescence to distal STEM career attainment in mid-adulthood. Grounded in the integrative frameworks of Social Cognitive Career Theory (SCCT) and Life Course Theory, our findings illuminate the complex, dynamic, and sequential processes through which a constellation of grade-7 socialization agents—parental educational encouragement, peer academic support, constructive school learning environment, and student self-esteem—collectively sculpt the developmental and growth trajectories of science performance throughout middle school, which in turn serve as critical mechanisms propelling students toward the attainment of a STEM degree and subsequent engagement in a STEM profession. This discussion synthesizes these findings, arguing that they not only fill a pronounced gap in the literature but also fundamentally challenge siloed intervention approaches, compelling a paradigm shift toward a more holistic, ecological, and temporally sensitive understanding of STEM development as a long-term continuum.

6.1. Theoretical and Empirical Advancements: From Static Snapshots to Dynamic Processes

The most profound contribution of this study lies in its successful operationalization of science learning in middle school as a dynamic, growth-oriented process rather than a static outcome, which in turn lays the foundation for long-term STEM development. Prior research has often treated academic performance as a cross-sectional predictor, inadvertently neglecting its malleable and evolving nature (Arens et al., 2017; Pekrun et al., 2017; Yeung & Igarashi, 2025). By employing Latent Growth Curve Modeling (LGCM), we captured the essence of science learning as a dual-faceted construct: the initial level (intercept) representing foundational preparedness at the onset of middle school, and the growth rate (slope) indicative of knowledge acquisition and skill development across this critical formative period. The finding that both the intercept and slope of middle school science achievement are significant predictors of earning a college STEM degree provides robust validation for this dynamic approach. It underscores that while a strong start is advantageous (reflecting prior preparation and capital), the capacity for growth during middle school itself is equally vital for long-term STEM success. This directly addresses the theoretical proposition of Life Course Theory that early trajectories cast a long shadow (Pallas, 2003), and aligns with SCCT’s emphasis on the development of performance efficacy and outcome expectations over time (Lent et al., 1994; Lent et al., 2018).

Furthermore, the negative correlation between the intercept and slope factors suggests a potential ceiling effect or differential investment of resources, wherein students who begin with high proficiency may experience less dramatic growth, a nuance that single-point assessments would completely obscure (Grimm et al., 2017). This finding invites a more sophisticated discussion about educational equity and targeted support for science and STEM education (Ayuso et al., 2022; Larkin & Lowrie, 2022), suggesting that interventions must be tailored not only to boost initial levels for some but also to sustain growth trajectories for all, especially high-achievers who might otherwise plateau. This finding robustly validates the dynamic approach central to our integrated theoretical framework, demonstrating that the ‘long shadow’ of early trajectories (Life Course Theory) is cast through the development of academic competencies and expectations (SCCT).

6.2. The Synergistic Yet Differential Power of Socialization Agents

Our results provide compelling evidence for the collective impact of multiple socialization agents, yet they also reveal a clear hierarchy and differentiation in their influences, offering nuanced insights for theory and practice. The robust effect of parental educational encouragement—the strongest predictor of both the initial level and growth of science achievement— reaffirms the primacy of the family context. Parents act as the primary architects of educational values, transmitting expectations and providing the emotional and instrumental scaffolding that fosters academic resilience and ambition (Boonk et al., 2018; Dotterer, 2022). Its significant direct effect on STEM degree attainment, even after accounting for science trajectories, points to its enduring role in shaping educational choices and persistence throughout the educational pipeline, likely through mechanisms of vicarious learning, social persuasion, and financial and emotional support during the challenging transition to higher education (Lent & Brown, 2013).

The significant role of peer academic support underscores the escalating importance of the peer microsystem during early adolescence. Peers provide a critical normative context, serving as sources of collaborative learning, academic motivation, and shared identity around educational pursuits (Altermatt, 2019; Gong et al., 2022). Its significant effect suggests that pro-academic peer networks create an environment where engagement with challenging subjects like science is valued and supported. Interestingly, while constructive school learning environment and student self-esteem were significant predictors of the initial science level, their influence on the growth trajectory was more selective. Only the school environment predicted growth, highlighting the unique role of authoritative structures—teachers and schools—in facilitating progressive mastery of learning. Teachers, as subject-matter experts, are uniquely positioned to provide the formal instruction, corrective feedback, and intellectual challenge necessary to navigate the increasing complexity of scientific concepts (Kloser et al., 2018; Newman et al., 2015). The finding that self-esteem did not directly predict growth, while it bolstered initial performance, suggests a distinction between global self-worth and domain-specific self-efficacy. It implies that while a positive self-concept provides general psychological capital, sustained growth in a specific, challenging domain like science may depend more on the cultivation of academic self-efficacy—a construct more directly tied to mastery experiences and verbal persuasion from authorities like teachers and parents (Perinelli et al., 2022; Scheeren, 2022). This differentiation is a crucial insight: a constructive school learning environment provides the specific, mastery-oriented experiences that translate general self-esteem into domain-specific competence and growth.

6.3. Middle School Science Trajectories: The Pivotal Mediating Mechanism

The study’s core contribution is the empirical demonstration that middle school science trajectories act as the key mediating mechanism through which the early adolescent socialization context influences adult STEM outcomes. The significant mediation effects (H4, H5) confirm that the influence of parents, peers, school, and self-esteem on ultimately working in a STEM field is largely, though not entirely, channeled through their success in fostering science proficiency during the critical “pre-action planning phase” (Lent & Brown, 2013; Lent et al., 1994). This mediation is not merely statistical; it represents a real-world process of capital conversion. The social and personal capital provided by the grade-7 socialization agents is effectively converted into academic capital—embodied in improving science scores—which then becomes the currency for accessing further educational opportunities (e.g., advanced placement courses, selective colleges) and ultimately, STEM careers (Dotterer, 2022; English, 2017; Penprase, 2020).

This finding carries immense implications. It moves the focus of intervention earlier in the developmental timeline, arguing that efforts to diversify the STEM pipeline must intensify during middle school, not just at the college level. It suggests that enhancing science education in middle school is not an end in itself but a powerful lever for altering long-term life trajectories. Policies that improve teacher quality, curriculum rigor, and hands-on learning experiences in middle school science are, by extension, investments in the nation’s future STEM workforce.

6.4. The Sequential Pathway to STEM Careers: The Non-Negotiable Role of the Degree

The study further refines our understanding of the pathway by establishing a sequential mediation model: socialization agents → science trajectories → STEM degree → STEM career. The formidable effect of obtaining a STEM degree on career entry (β = 0.608, increasing odds by 83.67%) powerfully underscores that while middle school performance sets the stage, the college degree remains the critical gateway. This aligns with human capital theory, positing that the degree signals specific competencies and knowledge to employers (Almatrafi et al., 2017; Delaney & Devereux, 2022). It also suggests that the degree itself provides access to specialized networks, internships, and credentials that are often prerequisites for STEM professions (Ekmekci et al., 2019; Zhang, 2022). This finding warns against an over-optimistic belief that passion and early aptitude alone are sufficient; structural barriers persist and are critically influential, and the formal credential of a degree remains a powerful, though not insurmountable, filter. Therefore, interventions must support students not only to enter the STEM pathway but also to persist through the “action phase” of degree completion, addressing attrition factors in undergraduate STEM programs.

6.5. Demographic Nuances: Unpacking Inequality in the Stem Pipeline

Our findings on demographic covariates paint a complex picture of stratification within the STEM pipeline, reinforcing the need for targeted, equity-focused policies. The advantage observed for male students in initial science development and STEM degree attainment is consistent with a body of literature on gender stereotypes and gender gaps in early STEM engagement (Friedman-Sokuler & Justman, 2016). However, the counterintuitive finding that women were more likely to be in STEM professions in adulthood is a critical nuance. This likely reflects the broad definition of STEM in this study, which includes life, health, and medical sciences—fields where female representation is historically higher (Wright et al., 2017). It suggests that women who persist through the pipeline and obtain a STEM degree may find more welcoming environments or greater opportunities in these specific sub-fields. This highlights the danger of treating “STEM” as a monolith and calls for more granular analyses of sub-disciplines. The significant advantages associated with a two-parent family structure point to the role of economic and social resources—what Bourdieu would term familial capital—in providing stability, academic support, and perhaps access to enrichment activities that foster early science interest (Heard, 2007; Pribesh et al., 2020).

Most striking are the persistent ethnic disparities. The underperformance of Hispanic, African American, and Native American students compared to their Asian peers in middle school science, and their lower likelihood of obtaining a STEM degree, underscores systemic inequities. These likely stem from a confluence of factors including unequal access to high-quality STEM instruction, underlying socioeconomic disparities, stereotype threat, and a lack of role models (Bottia et al., 2021; Miller & Pearson, 2012). The outperformance of Asian students, even when controlling for socialization agents, points to the potent influence of unmeasured cultural factors, such as specific parental expectations around educational achievement and the high value placed on STEM professions within some Asian communities (Kang et al., 2021; Liu & Xie, 2016). The finding that White students were also less likely than Asians to obtain a STEM degree, despite similar middle school performance, further complicates the narrative and warrants investigation into differences in choice architecture, perceived fit, or non-academic factors influencing major selection.

6.6. Policy and Intervention Implications: Toward a Synergistic Ecology of Support

The collective and sequential nature of our findings demands a radical rethinking of educational policy and intervention strategies. Siloed programs—focusing only on parents, or only on teacher training, or only on curriculum—are inherently limited. Our results argue compellingly for multi-systemic, coordinated interventions that simultaneously target the family, peer group, school, and individual student to create a coherent and reinforcing pro-STEM ecology.

• For Parents and Families: Programs should equip parents, especially those in underrepresented communities, with strategies to provide effective educational encouragement, moving beyond general support to specific science-oriented advocacy. This includes guiding them on how to discuss science and math positively and link them to real-world careers; monitor and discuss their child’s science progress; express clear and high educational expectations, particularly regarding college attendance; and actively help their children navigate educational planning for STEM pathways.
• For Schools and Teachers: Investment in professional development for middle school science teachers is paramount to create the supportive and rigorous environment that fuels growth. This includes training in inquiry-based and hands-on pedagogical methods that make science engaging. Furthermore, schools should deliberately structure collaborative, project-based learning to foster pro-academic peer networks and make science engagement socially rewarding. School policies should also prioritize creating a positive school climate characterized by caring teacher–student relationships and high academic standards, as our measure of constructive school learning environment directly predicted science growth.
• For Practitioners: Counselors and educators should use dynamic assessments of science trajectories (initial level and growth) to identify students who, despite a low initial starting point, show strong growth potential, and target them with enrichment resources. Conversely, high-achievers who may be at risk of plateauing need to be challenged to maintain their growth trajectory.
• For Policymakers: Funding priorities should emphasize the middle school years as a critical intervention point. Policies should encourage the integration of family and community engagement into school STEM initiatives, breaking down the walls between these systems. This could include funding for family STEM nights, community-school partnerships, and grants for schools to implement the multi-level supports described above.”

6.7. Limitations, Future Directions, and Concluding Synthesis

While this study provides a robust longitudinal map, it is not without limitations. The use of LSAY data, though a strength in terms of longevity, means the experiences captured are from a cohort that navigated the educational system over a decade ago. The dramatic evolution of technology, the increased emphasis on computational thinking, and the shifting landscape of STEM careers necessitate replication with contemporary longitudinal data (Larkin & Lowrie, 2022; Penprase, 2020). Second, the modeling assumed a unidirectional influence from socialization agents to science achievement. Future research should employ cross-lagged panel models to explore the likely reciprocal relationships, such as how improving science performance might elicit more parental encouragement and peers’ support (Arens et al., 2017; Yang et al., 2023). Third, while we focused on science, the integrated nature of academic achievement suggests that trajectories in mathematics are equally critical and their interplay with science trajectories should be examined (Bleeker & Jacobs, 2004; Kohen & Nitzan, 2022). Fourth, expanding the growth modeling to include high school science trajectories would provide an even more complete picture of the academic pathway (Bottia et al., 2018; Starr et al., 2022). Fifth, the LSAY cohort entered middle school in 1987. Over the subsequent three decades, the nature of socialization agents has likely evolved significantly due to the rise of the internet, social media, and shifting educational policies. While the fundamental mechanisms identified in this study are likely still relevant, the relative influence of peers (now often including online networks) or the ways parents and schools provide encouragement may have changed. Future research should replicate the models vindicated in the current study with contemporary longitudinal data to map these potential changes and update our understanding of the socialization ecology for today’s students. Finally, the dichotomous operationalization of STEM outcomes, while practical, masks important variation within STEM fields (Miller & Pearson, 2012; Wright et al., 2017). Future work should strive to understand the pathways into specific, high-demand sub-fields like computer science versus the life sciences.

In sum, this study moves beyond simplistic correlations to delineate a dynamic, ecologically embedded, and sequential pathway to STEM success. It empirically establishes that the roots of the STEM workforce are planted not in college only, but in the collaborative soil of the middle school environment, nurtured by parents, peers, teachers, and the students’ own sense of self. The journey from a seventh-grade classroom to a STEM profession is a marathon, not a sprint, with relay points at the end of middle school and the completion of a college degree. The findings serve as a powerful call to action: to cultivate a diverse and robust STEM pipeline, we must invest in creating synergistic, supportive ecologies that empower all adolescents to build and sustain strong trajectories of scientific learning from the very beginning of their educational careers. The middle school years are not merely a preparation for high school; they are the foundational bedrock upon which future innovation and economic competitiveness are built.

7. Conclusions

This study has successfully charted a critical and previously underexplored pathway to STEM success, demonstrating that the foundation for a career in science, technology, engineering, and mathematics is firmly established not in college, but during the formative middle school years. By integrating the theoretical frameworks of Social Cognitive Career Theory and Life Course Theory, we have provided robust, longitudinal evidence that the journey to STEM is a continuous process, shaped by a synergistic ecology of support and marked by dynamic academic growth. The key importance of this research lies in its holistic and dynamic approach. It moves beyond examining isolated factors to reveal how a constellation of socialization agents—parental educational encouragement, peer academic support, a constructive school learning environment, and student self-esteem—collectively operates in seventh grade to initiate a positive cascade. This collective context directly fosters both a higher starting point and a steeper growth trajectory in science achievement throughout middle school. These trajectories are not merely academic metrics; they are the key mediating mechanisms that convert early social and personal capital into the academic capital necessary for long-term achievement. Ultimately, this process culminates in the attainment of a STEM degree, which serves as the decisive gateway to entering a STEM profession in adulthood.

The implications of these findings are substantial and demand a concerted shift in educational policy and practice. To cultivate a diverse, robust, and innovative future STEM workforce, efforts must be prioritized and intensified at the middle school level. This necessitates a move away from siloed interventions and towards integrated, multi-systemic strategies that simultaneously engage families, peers, educators, and the students themselves. In conclusion, this research provides a compelling evidence base for re-envisioning STEM education as a long-term, ecological continuum. It affirms that empowering all adolescents to build and sustain strong trajectories of scientific learning from the very beginning of their educational careers is not just an educational objective, but a socioeconomic imperative. The middle school years are the foundational bedrock upon which future innovation is built. By fostering a synergistic pro-science context during this critical period, we can effectively pave the way for long-term STEM success, ensuring a diverse and capable generation is ready to meet the scientific and technological challenges of the future.