Advancing Saudi Vision 2030 for Sustainable Development: Modeling Influencing Factors on Adolescents’ Choice of STEM Careers Using Structural Equation Modeling, with a Comparative Analysis of Bahrain and Singapore

Abstract

Science, technology, engineering, and mathematics (STEM) are crucial for economic development and play a significant role in achieving sustainable development goals. Despite this, there is a shortage of skilled STEM professionals and a declining interest in STEM education and careers. The Saudi Vision 2030 goal of economic diversification and sustainable development aims to transform Saudi Arabia into a knowledge-based economy driven by innovation and sustainability. This study investigates factors influencing adolescents’ attitudes toward STEM careers in Saudi Arabia, with comparative insights from Bahrain and Singapore. Structural equation models (SEM) were constructed for each country to analyze the influence of scientific self-concept, school belonging, and teacher effectiveness on students’ choices of science careers. Mediation analysis examined the interest and value of science as mediators in these relationships. Confirmatory factor analysis was conducted to validate model constructs before building SEM models. Data from TIMSS 2019 for eighth-grade students was used to develop model constructs based on relevant items from the student questionnaire. Findings reveal that students’ interest in and value of science significantly influence career decisions, with self-concept and teacher engagement playing crucial roles. Teacher effectiveness had the strongest impact on science interest in Saudi Arabia and Bahrain, while self-concept was most influential in Singapore. These results highlight the importance of fostering teacher engagement and self-concept to encourage students’ career paths in science. To support this, Saudi Arabia should enhance teacher training programs by integrating mentorship, active learning strategies, and technology driven instruction to improve student engagement. Adopting Singapore’s blended learning model can foster self-confidence and independence in STEM education, while hands-on learning and career exposure programs can strengthen students’ self-concept and long-term commitment to STEM fields. Additionally, expanding extracurricular STEM initiatives and industry partnerships will help connect classroom learning to real-world applications. By aligning STEM education reforms with these insights, Saudi Arabia can cultivate a skilled workforce that supports its economic transformation under Vision 2030.

1. Introduction

Science contributes to prosperity, well-being, and increased opportunities in nearly every aspect of our lives. Throughout history, science has played a significant role in human progress. As nations confront the challenges of sustainable development, fostering a robust pipeline of talent in science, technology, engineering, and mathematics (STEM) has become essential. Therefore, the Kingdom of Saudi Arabia has announced long-term goals as part of Saudi Vision 2030. The objective is to replace the country’s oil-based economy with a knowledge-based one contributing to sustainable development [1,2]. To achieve the objectives of Vision 2030 and promote the knowledge economy, the government is prioritizing investment in the STEM fields, aiming to produce more highly skilled professionals in STEM who are capable of driving innovation and supporting sustainable growth. Thus, the STEM field is being placed in educational and policy circles as a potential instrument of success more than ever before [3]. This is increasing the number of jobs in the STEM field, which is associated with the economic development of a country. Therefore, it is a national priority to prepare the next generation of workers and leaders with STEM skills. However, there has been a declining interest in STEM-related subjects and a decreasing motivation to pursue a career in this field. As a result, the number of STEM professionals worldwide may decline [4]. This raises significant concerns about the shortage of STEM professionals trained internationally [3]. For example, as reported by the Ministry of Education Education Statistics Center (2021), Saudi Arabia graduated 256,363 students from colleges and universities during the 2017–2018 academic year. Of these, only 6.9% were in the fields of science, mathematics, and statistics, 6.3% in technology, and 7.9% in engineering and building fields [5].

On the other hand, it is important to highlight that the OECD’s PISA 2022 Results indicate a substantial decline in student performance in mathematics, reading, and science across OECD countries from 2003 to 2022, raising concerns regarding students’ preparedness for careers in STEM [6]. Moreover, Kennedy et al. [7] assert that the declining trends in science education enrollment observed in Australia are also evident in various other countries worldwide. In line with this global trend, students from Saudi Arabia and other Gulf Cooperation Council (GCC) countries have declined in international assessments such as the Trends in International Mathematics and Science Study (TIMSS). Consequently, this decline is likely to intensify the lack of interest in STEM-related careers, presenting a significant barrier to efforts aimed at fostering sustainable development [8].

It is challenging to encourage adolescents to pursue STEM careers; therefore, it is necessary to analyze the selection process and examine the factors that influence students’ attitudes toward STEM career paths. Saudi Arabia seeks to align youth career goals with its sustainability objectives. Thus, this study seeks to explore the alignment with social cognitive career theory (SCCT) by examining factors that predict students’ interest in STEM-related careers in Saudi Arabia, with comparisons to Bahrain and Singapore. Singapore is selected for comparison due to its exceptional performance in the TIMSS assessments, where it achieved the highest average scores in science among all participating countries, with a score of 608 points, 108 points above the TIMSS scale center point [9]. Bahrain is included for its cultural and educational similarities with Saudi Arabia and its high achievement within the Gulf region, recording an average science score of 486 points in TIMSS 2019 [9]. By comparison, Saudi Arabia’s average science score for eighth-grade students in TIMSS 2019 was 431 points [9].

Consequently, it is worthwhile to investigate the factors influencing Singaporean and Bahraini students’ attitudes toward STEM career paths and to compare these factors with those affecting Saudi students.

There are a lot of different factors that may affect a student’s decision choice of scientific career, such as gender, family background, culture, social support, outcome expectations, learning experience, self-efficacy, and socio-economic background. The main purpose of this research is to construct three models that illustrate how various factors impact students’ attitudes toward STEM careers in Saudi Arabia, Bahrain, and Singapore, as well as the interconnection between these factors. Analyzing the factors influencing students’ attitudes in Saudi Arabia in comparison to those in Bahrain and Singapore could provide insights into the decline in Saudi students’ interest in STEM careers and propose potential solutions aligned with sustainable development goals based on the experiences of other nations.

The study is framed around the following research questions:

• How do teachers, belonging to school, and self-concept influence adolescents’ interest in science and adolescents’ perceived value of science?
• How do adolescents’ interest in and perceived value of science influence their career expectations, particularly in pursuing a STEM career?
• How do teachers, belonging to school, and self-concept influence adolescents’ career expectations in STEM through their interest in and value of science?

2. Literature Review

SCCT, as provided by Lent et al. [10], is a career theory that focuses on personal interests and human improvement. Hence, SCCT can be used to analyze the factors influencing the career choice. It elucidates three interconnected aspects of career development, namely, interest, choice, and career performance and success. Additionally, this theory considers various significant factors such as culture, gender, values, life events, and other circumstances that could impact career development and choices. As well, health status, learning experience, self-efficacy, expectations, interests, personal inputs, and environment could affect career choices [10]. Furthermore, SCCT has been a major source of the general viewpoint expressed in the literature on students’ career decisions. Several studies have employed the SEM technique with SCCT to illustrate the factors that affect career choices.

A review of studies employing SCCT reveals common patterns in factors influencing STEM career aspirations. For example, Qi’s comparative study [11] between the United States and Singapore found that self-efficacy played a significant role in shaping career aspirations, though its impact varied between cultural contexts. Similarly, Badri et al. [12] highlighted the importance of teacher support and classroom experiences in motivating students toward STEM careers in Abu Dhabi. These findings align with Kang’s research [13] in Finland, which emphasized the role of positive learning experiences in enhancing self-efficacy and fostering interest in science careers. Furthermore, Nakamura et al. [14] found that self-efficacy and outcome expectations in science learning positively influence STEM career aspirations, leading to increased student participation when comparing the influencing factors across Japan, Korea, Taiwan, and the U.S. Despite the differences in geographic and educational settings, these studies collectively suggest that self-efficacy and support systems (such as teachers and family) are universal predictors of STEM aspirations.

However, there are conflicting findings regarding the relative impact of self-efficacy versus external influences. While self-efficacy is emphasized in Qi’s [11] study, Kang’s [13] research indicates that broader educational structures, such as inquiry-based learning, may be more significant determinants of career interest. Similarly, Mau and Li [15] and Blotnicky et al. [16], demonstrated that early exposure to STEM opportunities, mentorship, and institutional support had a more profound effect, challenging the notion that self-efficacy alone is the primary driver. In addition, Zhao et al. [17] conducted an exploratory SEM study in Indonesia, identifying teaching quality, learning environment, student engagement, and perceived relevance of STEM content as key determinants of student satisfaction. Beside these studies, El-Dief and El-Dief [18] studied Saudi university students pursuing hospitality and tourism management and found that self-interest, outcome expectations, work environment, and industry experience were key predictors of career commitment. Their study also highlighted the influence of workforce nationalization and social status on students’ career choices, demonstrating how economic and policy factors interact with personal aspirations.

Furthermore, research by Alsalamat [19] in Saudi Arabia underscores the implementation challenges of STEM education due to inadequate resources and insufficient teacher training. The findings underscore the need for targeted professional development programs and better resource allocation to support effective STEM integration in Saudi classrooms. In addition, Al-Zahrani [20] in his study confirmed that novice teachers in GCC countries report more weaknesses than strengths in their teacher education programs (TEP), highlighting the need for significant reforms. A systematic review of studies from 2010 to 2022 identified key concerns, including the theory-practice gap, non-culturally responsive curricula, and insufficient alignment between coursework and real-world teaching demands. While practicum courses were seen as beneficial, novice teachers emphasized the necessity of improving course quality, bridging the gap between schools and training programs, and developing essential work-related skills. To address these challenges, policymakers, institutions, and educators should conduct comprehensive evaluations involving all stakeholders, including TEP graduates, to ensure continuous program improvement. Additionally, future research should expand on novice teachers’ experiences to ensure that teacher training in the GCC meets evolving educational needs.

Recent studies have explored various aspects of STEM education, shedding light on key factors influencing learning outcomes and engagement. Spagnolo et al. [21] highlight the role of storytelling as an interdisciplinary approach to teaching STEM, emphasizing its potential to bridge the gap between science and the humanities by enhancing conceptual understanding and student engagement. Bianca and Spagnolo [22] investigate gender differences in perceived mathematical difficulty, revealing that female students often report higher levels of difficulty, which may be influenced by self-efficacy, societal stereotypes, and learning experiences, suggesting the need for gender-sensitive teaching strategies. These studies collectively emphasize the importance of innovative teaching methods, inclusivity, and cognitive research in improving STEM education and fostering student success.

While SCCT provides a valuable framework for understanding career aspirations, the literature highlights a complex interplay between self-efficacy, external support, and broader systemic factors. Differences in cultural contexts, educational structures, and resource availability influence career choices, emphasizing the need to support STEM engagement.

These studies indicate that psychological and motivational factors influencing students’ attitudes toward science may differ from one nation to another. Therefore, comparing Saudi Arabia with Bahrain and Singapore in this study may highlight different perspectives on the factors influencing students’ career decisions.

3. Materials and Methods

3.1. Theoretical Framework

The theoretical framework is built using the dimensions of SCCT. TIMSS data are big and contain many items that express multiple factors. From TIMSS, the items that correspond to the factors in SCCT are selected to form the theoretical framework. The factors value of science, interest, self-concept, teacher effectiveness, and belonging to school are selected to study how they affect the student’s choice of future job.

Figure 1 describes the theoretical model. The figure shows five factors (latent variables) and one observed variable reflecting future job choices. From Figure 1, the dependent variable (endogenous variable) is a future job. Self-concept (SC), teacher effectiveness (teacher), and belonging to school (school) are independent variables (exogenous variables). Interest in science and value of science are mediator variables that are simultaneously considered dependent and independent variables. The independent variables can affect future jobs directly or indirectly through the mediator variables. The concepts associated with these latent variables are as follows:

Interest: This construct refers to the intrinsic enjoyment or pleasure derived from engaging in a specific activity [23]. It emphasizes the role of positive affect and personal enjoyment in motivating individuals to pursue and persist in activities they find engaging and rewarding. Interest, along with positive emotional responses, plays a significant role in influencing the pursuit of enjoyable and stimulating tasks [24,25].

Value of Science: This factor reflects students’ perceptions of the relevance and importance of science in achieving their future goals. Instrumental incentives, such as the belief that excelling in science is necessary for gaining admission to a desired university, exemplify how students view science as a means to achieve broader academic and career objectives.

Self-Concept: Rooted in expectancy-value theory, self-concept in science refers to students’ beliefs in their scientific abilities [26]. An individual’s confidence in their capacity to succeed in science is central to their self-concept and can influence their attitude toward pursuing science-related careers. For example, a student’s statement, such as “Science is not my strength”, reflects their self-perception of competence in science.

Belonging to School: This factor pertains to the emotional connection a student feels toward their school environment, which is shaped by the school’s disciplinary climate and the student’s sense of safety and acceptance [27]. A positive sense of belonging is fostered when students feel supported, while it can be undermined if they perceive the school environment as unjust or unfriendly [27]. Research has shown that a high level of school belonging is linked to positive academic outcomes, increased motivation, and better psychological well-being [28]. The disciplinary environment significantly impacts students’ attachment to their school, as feeling safe and treated fairly reinforces their sense of belonging. Conversely, if students view school regulations as unfair, their connection to the school may weaken [27]. Schools play a pivotal role in fostering this sense of belonging by cultivating positive relationships, maintaining a secure environment, and implementing just and transparent disciplinary policies.

Teacher Effectiveness: This variable captures students’ perceptions of their science teachers’ ability to effectively communicate content, engage students, and provide necessary resources and support in the science classroom. Teacher effectiveness is considered a critical predictor of student achievement, which, in turn, can influence students’ motivation to pursue STEM careers. This factor illustrates the feelings about one’s science teacher. Students’ perceptions of professors’ ability to communicate engaging subjects, standards, and resources in scientific classrooms serve as a measure of the effectiveness of their instruction [29]. Additionally, it is an important predictor of students’ achievement and, thus, may influence their career choices.

These latent variables are crucial in understanding the factors that influence students’ attitudes toward science and their career choices, highlighting the role of personal beliefs, external perceptions, and the learning environment.

3.2. Sampling Strategy

TIMSS is conducted by the International Association for the Evaluation of Educational Achievement (IAEEA), employs a comprehensive and statistically rigorous sampling process to ensure the data is representative and reliable across participating countries. The target population consists of students in grades 4 and 8, who are assessed in mathematics and science. For eighth-grade students, TIMSS targets the grade corresponding to eight years of formal schooling starting from the International Standard Classification of Education Level 1, with the mean age at the time of testing being at least 13.5 years.

The sampling frame is built using national education statistics or school directories to create a list of eligible schools. Stratified random sampling is then applied, where schools are divided into different strata (e.g., by region or school type) and schools are randomly selected from each stratum to ensure diversity and representativeness. TIMSS uses multistage sampling, where schools are first selected, and then students within those schools are randomly chosen. This ensures that the sample captures a wide range of student backgrounds, including urban, rural, public, and private school students. The sample size is designed to be large enough to provide reliable data that can be generalized to the entire population of students in each country.

Once data is collected, weighting is applied to adjust for any potential biases, such as nonresponse or differences in the probability of selection between different groups of students or schools. This helps correct for underrepresentation or overrepresentation of certain groups, ensuring that the final sample reflects the demographic makeup of the population. In addition, non-response adjustments and imputation techniques are used to account for missing data, with the goal of minimizing bias from students or schools that did not participate. Post-stratification may also be employed to further adjust the sample to match the known population parameters.

The process also includes oversampling of smaller or underrepresented groups, such as rural schools or particular demographic categories, to ensure that they are adequately represented. Throughout the data collection process, strict quality control procedures are in place, including training for assessors, field supervision, and checks to ensure consistency and accuracy. Instruments such as tests and surveys are calibrated to ensure that they effectively measure the intended knowledge and skills of students. These comprehensive steps collectively help to ensure that the TIMSS sample is unbiased, highly representative, and produces valid data to analyze global student performance in mathematics and science.

TIMSS, conducted every four years, enables the monitoring of educational trends over time. Data is collected through standardized tests and contextual surveys, offering insights into the factors that influence student performance across various countries.

Data is gathered through four different questionnaires targeting students, teachers, school principals, and parents. These questionnaires provide detailed information on learning environments, instructional practices, and school contexts. TIMSS, conducted every four years, enables the monitoring of educational trends over time. Data is collected through standardized tests and contextual surveys, offering insights into the factors that influence student performance across various countries [30,31,32].

3.3. Data and Variables

The TIMSS 2019 science data set for Saudi Arabia, Singapore, and Bahrain is used in this study. The TIMSS international database, provided by IAEEA, is publicly available on its website [33]. The sample group consists of eighth-grade students with an average age of 13.9 years. The school sample, sample distribution, and sample size for each country are presented in

Table 1. Sample Size.

Country	Number of School	Male Students	Female Students	Total Sample Size
Saudi Arabia	222	2791	2884	5675
Bahrain	185	3022	2703	5725
Singapore	187	2487	2366	4853

.

The variables chosen from the student background questionnaire represent the study factors. Thirty-three items were chosen to determine students’ attitudes and perspectives about learning science. Each item has four potential responses: “disagree a lot”, “disagree a little”, “agree a little”, and “agree a lot”, coded from 4 to 1, respectively (four-point Likert scale) [34]. All 33 items are reverse scored except for negative attitudes like “science is boring”, using the SPSS Program before analyzing [35]. Subsequently, greater values reflect greater and more positive attitudes toward science. The response variable is named future job and contains one item, “How much do you agree that you would like a job that involves using science?”.

Table 2. The selected variables and corresponding factors.

Latent Variables	Variable Description	Variable Name
Interest	“I enjoy learning science.”	Int1
	“I wish I did not have to study.”	Int2
	“Science is boring.”	Int3
	“I learn many interesting.”	Int4
	“I like science.”	Int5
	“I look forward to learning.”	Int6
	“Science teaches me how things.”	Int7
	“I like to conduct science experiments.”	Int8
	“Science is one of my favorite subjects.”	Int9
Self-Concept	“Science is more difficult for me.”	SC1
	“Science is not one of my strengths.”	SC2
	“Science is harder for me than any other subject.”	SC3
	“Science makes me confused.”	SC4
Value of Science	“I think learning science will help.”	VS1
	“I need science to learn other school subjects.”	VS2
	“I need to do well in science to get into the <university> of my choice.”	VS3
	“I need to do well in science to get the job I want.”	VS4
	“It is important to learn about science to get ahead in the word.”	VS5
	“Learning science will give me more job opportunities when I am an adult.”	VS6
	“My parents think that it is important that I do well in science.”	VS7
	“It is important to do well in science.”	VS8
Belonging to School	“I like being in school.”	BS1
	“I feel safe when I am at school.”	BS2
	“I feel like I belong at this school.”	BS3
	“I am proud to go to this school.”	BS4
Teacher Effectiveness	“I know what my teacher expects me to do.”	TE1
	“My teacher is easy to understand.”	TE2
	“My teacher has clear answers to my questions.”	TE3
	“My teacher is good at explaining science.”	TE4
	“My teacher does a variety of things to help us learn.”	TE5
	“My teacher links new lessons to what I already know.”	TE6
	“My teacher explains a topic again when we don’t understand.”	TE7
Future Job	“I would like a job that involves using science.”	Job

shows the items for each factor.

3.4. Statistical Methods

Descriptive statistics and Structural SEM are employed in this study to analyze the factors influencing students’ career choices in STEM fields. SEM, a multivariate statistical technique also known as causal modeling [36], allows researchers to analyze the associations between multiple dependent and independent variables. Furthermore, the mediation analysis can aid in uncovering the factors that directly and/or non-directly impact student attitudes toward selecting careers in STEM. This approach is particularly valuable in identifying mediating effects, where certain factors may not exert a direct influence but contribute indirectly to students’ career aspirations. SEM consists of two key components: the measurement model and the structural model. The measurement model, also referred to as Confirmatory Factor Analysis (CFA), is used to assess whether the observed variables reliably represent their underlying latent constructs [37]. CFA requires evaluation from multiple perspectives, including multicollinearity, unidimensionality, reliability, and validity, and is assessed using various model fit indices. By validating the measurement model, CFA ensures that only the most reliable indicators are retained, enhancing the construct validity and measurement accuracy of the model. The structural model, on the other hand, examines causal relationships between constructs by integrating the measurement and structural components into a unified SEM framework. Unlike traditional statistical methods that rely on composite scores or calculated variables, SEM provides a more precise estimation of latent variables, allowing for a comprehensive analysis of the relationships between factors affecting students’ STEM career choices. The results of these statistical analyses will be presented in the next section.

4. Results

4.1. Descriptive Statistics

The descriptive statistics for variables concerning eighth-grade students in Saudi Arabia, Bahrain, and Singapore were calculated.

Table 3. Descriptive Statistics.

Variables	Frequency	Job
		Saudi Arabia		Bahrain		Singapore
		Agree	Disagree	Agree	Disagree	Agree	Disagree
Interest	Very Much Like Learning Science	2076 (88.3%)	274 (11.7%)	2072 (86.1%)	335 (13.9%)	1600 (89.3%)	192 (10.7%)
	Somewhat Like Learning Science	1312 (65.7%)	685 (34.3%)	1351 (61.0%)	864 (39.0%)	1452 (61.4%)	914 (38.6%)
	Do Not Like Learning Science	204 (30.5%)	464 (69.5%)	272 (31.4%)	595 (68.6%)	160 (23.4%)	523 (76.6%)
SC	Very Confident in Science Due to Strong Self-Concept	1476 (81.5%)	335 (18.5%)	1558 (78.9%)	416 (21.1%)	899 (87.4%)	130 (12.6%)
	Somewhat Confident in Science due to Self-Concept	1336 (64.7%)	728 (35.3%)	1371 (59.1%)	950 (40.9%)	1686 (67.4%)	817 (32.6%)
	Not Confident in Science due to Weak Self-Concept	779 (68.6%)	356 (31.4%)	762 (64.0%)	428 (36.0%)	630 (47.9%)	685 (52.1%)
Value	Strongly Value Science	2637 (94.9%)	143 (5.1%)	2692 (94.2%)	166 (5.8%)	1906 (94.5%)	110 (5.5%)
	Somewhat Value Science	931 (54.7%)	771 (45.3%)	964 (49.8%)	971 (50.2%)	1285 (55.7%)	1022 (44.3%)
	Do Not Value Science	43 (7.7%)	516 (92.3%)	51 (7.1%)	665 (92.9%)	25 (4.8%)	500 (95.2%)
School	High Sense of School Belonging	1721 (79.1%)	454 (20.9%)	1599 (75.3%)	525 (24.7%)	999 (74.1%)	349 (25.9%)
	Some Sense of School Belonging	1497 (68.0%)	704 (32.0%)	1624 (64.7%)	885 (35.3%)	1828 (64.6%)	1000 (35.4%)
	Little Sense of School Belonging	343 (57.6%)	252 (42.4%)	456 (54.2%)	385 (45.8%)	388 (57.8%)	283 (42.2%)
Teacher	High Teacher Clarity	2745 (81.2%)	634 (18.8%)	2493 (76.5%)	766 (23.5%)	1594 (78.3%)	441 (21.7%)
	Moderate Teacher Clarity	789 (56.0%)	619 (44.0%)	1081 (58.3%)	772 (41.7%)	1512 (59.7%)	1019 (40.3%)
	Low Teacher Clarity	68 (28.7%)	169 (71.3%)	118 (31.2%)	260 (68.8%)	107 (38.8%)	169 (61.2%)
Gender	Female	1911 (71.4%)	765 (28.6%)	1792 (67.5%)	861 (32.5%)	1433 (60.6%)	933 (39.4%)
	Male	1702 (71.9%)	665 (28.1%)	1915 (67.1%)	941 (32.9%)	1783 (71.8%)	699 (28.2%)

displays the descriptive statistics for the study variables. The tables provide the frequencies and percentages of each factor across the job variable.

The data presented in Table 3 reveals that the distribution of students across different categories is similar in the three countries. Notably, more than 50% of students in the three levels of their confidence in science, express interest in pursuing careers in science. In Saudi Arabia, 81.5% of students with high confidence in science are interested in science-related jobs, while 64.7% of those with medium confidence, and 68.6% of those with low confidence, also express support for such careers. Similar trends are observed in Bahrain and Singapore. The high percentages across all confidence levels in science suggest that confidence in science might not significantly influence students’ career preferences.

It appears that students with low confidence in science still exhibit a high interest in STEM careers across the three countries. While this may initially seem contradictory, Schunk [38] explains that confidence and interest are distinct factors in academic and career choices. Confidence reflects a student’s self-perception of their ability in a subject, while interest pertains to the intrinsic desire to engage with that subject. It is entirely possible for students to be highly interested in STEM fields without feeling confident in their abilities. Additionally, external influences like exposure to STEM role models or encouragement from teachers and parents can further fuel this interest [39,40]. Early engagement in STEM-related activities also plays a critical role in cultivating a passion for these careers, even among students with low self-confidence [41]. The prestige and societal importance of STEM, reinforced by economic and educational reforms, further drive students’ aspirations, regardless of their confidence in science [40,42]. This combination of factors explains why students with low confidence in science may still pursue STEM careers.

Regarding the sense of belonging to school, more than 50% of students in the three levels of belonging to their school, express a desire to pursue careers in science. In Saudi Arabia, 79.1% of students with a strong sense of school belonging support science careers, compared to 68.0% among those with moderate belonging, and 57.6% among those with a low sense of belonging. The high percentages across all levels of school belonging in all three countries indicate that this variable may not appear to have a substantial effect on career aspirations.

Additionally, the table shows that there is little variation in the percentage of male and female students who are interested in science-related jobs. In Saudi Arabia, for example, 71.9% of male students and 71.4% of female students express support for science careers. Similar trends are observed in Bahrain and Singapore, suggesting that gender may not play a significant role in shaping students’ career aspirations.

When examining the influence of interest in science, a noticeable difference emerges: students who express a strong interest in learning science are more likely to support science careers compared to those who are less enthusiastic. In Saudi Arabia, 88.3% of students who have a strong passion for science are inclined toward science jobs, while 65.7% of those with moderate interest and only 30.5% of students who dislike learning science support such careers. These findings suggest that interest in science may have a considerable impact on career choices.

Similarly, the value students place on science is another factor that might influence students’ career preferences. In Saudi Arabia, 94.9% of students who highly value science express interest in science-related careers, while only 54.7% of those who somewhat value science and just 7.7% of students who do not value science express the same interest. A similar pattern is observed in Bahrain and Singapore, indicating that the perceived value of science significantly affects students’ career aspirations.

Lastly, teacher effectiveness appears to play a role in students’ career preferences. In Saudi Arabia, 81.2% of students who report high teacher effectiveness express support for science jobs, while 56.0% of those who report moderate effectiveness and only 28.7% of students who perceive low effectiveness support these careers. This trend is consistent in Bahrain and Singapore, suggesting that the quality of science teaching may influence students’ career intentions.

Finally, while factors such as confidence in science, school belonging, and gender appear to have little impact on students’ interest in science careers, factors such as interest in science, the value placed on science, and the perceived effectiveness of science teaching seem to have a more pronounced influence. These relationships and influences will be examined through mediation analysis and SEM models. However, there are some differences, where the agreement percentages are generally higher for variables like the strong value of science. Hence, it is anticipated that value will have a significant impact on future career choices. In addition, gender differences are more pronounced in Singapore compared to Saudi Arabia and Bahrain, although they do not show significant gender differences.

4.2. Confirmatory Factor Analysis

Establishing a high-quality measure is necessary before utilizing SEM. So, two models were implemented by AMOS program [43], the measurement and the structural. Figure 2 shows the general CFA model for the three countries, Saudi Arabia, Bahrain, and Singapore. The models include five constructs: Interest, Value of Science (Value), Self-Concept (SC), Teacher Effectiveness (Teacher), and Belonging to School (School). The examination requirements of the CFA model are discussed in the following.

4.2.1. Multicollinearity

The constructs are considered not redundant if the correlation value between them is less than 0.90 [44]. The two-way arrows represent the correlations between the factors, which must not exceed 0.90 to avoid the issue of multicollinearity [45].

Table 4. Correlation Between Constructs.

Pair Constructs	Saudi Arabia	Bahrain	Singapore
Interest <-> SC	0.44	0.48	0.65
Interest <-> Value	0.71	0.71	0.65
Interest <-> School	0.37	0.34	0.30
Interest <-> Teacher	0.67	0.67	0.58
SC <-> Value	0.25	0.28	0.36
SC <-> School	0.15	0.15	0.13
SC <-> Teacher	0.24	0.28	0.31
Value <-> School	0.32	0.34	0.29
Value <-> Teacher	0.62	0.61	0.48
School <-> Teacher	0.35	0.39	0.32

shows the correlation between construct. From Table 4, the highest correlation in the Saudi Arabia model is observed between the interest and value variables, with a value of 0.71, while the lowest correlation is between SC and School, with a value of 0.15. For the Bahrain model, the highest correlation between the interest and value variables is slightly lower, at 0.69, with the lowest correlation between SC and School at 0.16. Similarly, in the Singapore model, the highest correlation is again between the interest and value variables, at 0.65. Additionally, there is a high correlation between the interest and SC variables, which is equal to the correlation between the interest and value variables. The lowest correlation is between SC and School, at 0.13. As all correlation values between the constructs are below 0.9, the initial model meets the correlation condition for all three models.

4.2.2. Goodness-of-Fit

To evaluate the goodness-of-fit (G.O.F) of a model, various values are utilized. Table 5 presents the values of the G.O.F. indices along with the corresponding levels of acceptance for each index. From Table 5, for the Saudi model, the RMSEA is 0.062, the PGFI is 0.762, and the PNFI is 0.812, indicating that the level of acceptability for these indices has been achieved. However, the CFI is 0.892, the GFI is 0.886, the TLI is 0.882, and the NFI is 0.887, all of which are slightly below the acceptable threshold and require improvement.

For the Bahrain model, the RMSEA is 0.055, the GFI is 0.910, the CFI is 0.929, the TLI is 0.923, the NFI is 0.926, the PGFI is 0.778, and the PNFI is 0.844, indicating that the acceptability criteria for all indices have been satisfied. However, it is noteworthy that the GFI value is close to the minimum acceptance threshold, indicating that further model improvements could enhance the fit.

For the Singapore model, the RMSEA is 0.061, the CFI is 0.919, the TLI is 0.909, the NFI is 0.912, the PGFI is 0.763, and the PNFI is 0.835, suggesting that the level of acceptability for these indices has been met. However, the GFI is 0.888, which is slightly below the acceptable level and needs enhancement.

The ${\chi }^2$/df value may be disregarded in all models, as in many research, because it is very sensitive to the size of the sample, which inflates when the sample size increases [46,47,48]. Numerous studies have demonstrated that sample size has an impact on G.O.F indices, particularly the chi-square value [47,48]. Owing to the study’s large sample size for all three models, the needed value for ${\chi }^2$/df could not have been obtained. As some indications do not reach the level of acceptance, the model needs to improve till the indicators attain the acceptance level.

Table 5. Model Fit Indices.

Name of Category	Name of Index	Level of Acceptance	Saudi Model	Bahrain Model	Singapore Model	Reference
Absolute Fit	RMSEA	≤0.07	0.062	0.055	0.061	[49]
	GFI	≥0.90	0.886	0.91	0.888
	${\chi }^2$/df	≤5.00	23.053	18.334	19.296
Incremental Fit	CFI	≥0.90	0.892	0.929	0.919	[49]
	TLI	≥0.90	0.882	0.923	0.909
	NFI	≥0.90	0.887	0.926	0.912
Parsimonious Fit	PGFI	≥0.05	0.762	0.778	0.763	[50]
	PNFI	≥0.05	0.812	0.844	0.835

Note. RMSEA: Root Mean Square Error; GFI: Goodness Fit Index; CFI: Comparative Fit Index; TLI: Tucker-Lewis fit index; NFI: Normed Fit Index; PGFI: Parsimonious Goodness-of-Fit Index; PNFI: Parsimonious Normed Fit Index.

Table 5. Model Fit Indices.

Name of Category	Name of Index	Level of Acceptance	Saudi Model	Bahrain Model	Singapore Model	Reference
Absolute Fit	RMSEA	≤0.07	0.062	0.055	0.061	[49]
	GFI	≥0.90	0.886	0.91	0.888
	${\chi }^2$/df	≤5.00	23.053	18.334	19.296
Incremental Fit	CFI	≥0.90	0.892	0.929	0.919	[49]
	TLI	≥0.90	0.882	0.923	0.909
	NFI	≥0.90	0.887	0.926	0.912
Parsimonious Fit	PGFI	≥0.05	0.762	0.778	0.763	[50]
	PNFI	≥0.05	0.812	0.844	0.835

Note. RMSEA: Root Mean Square Error; GFI: Goodness Fit Index; CFI: Comparative Fit Index; TLI: Tucker-Lewis fit index; NFI: Normed Fit Index; PGFI: Parsimonious Goodness-of-Fit Index; PNFI: Parsimonious Normed Fit Index.

4.2.3. Unidimensionality

Unidimensionality is a test used to assess whether each item within a factor meets the minimum standard for acceptable factor loadings [51]. Items with the lowest factor loadings in the latent construct are deleted one by one until all loadings exceed 0.5, as recommended by [52]. The measurement model is then retested until it meets the criteria for unidimensionality. In our analysis, several variables were found to have weak loadings. These weak variables are eliminated incrementally to improve the model’s quality, validity, and reliability.

Table 6. Loading Items.

Factor	Item	Saudi Loading	Bahrain Loading	Singapore Loading
Interest	Int1	0.78	0.78	0.85
	Int2	0.37	0.46	0.68
	Int3	0.48	0.54	0.66
	Int4	0.77	0.78	0.71
	Int5	0.86	0.89	0.90
	Int6	0.82	0.83	0.87
	Int7	0.57	0.69	0.62
	Int8	0.58	0.57	0.49
	Int9	0.82	0.84	0.87
SC	SC1	0.72	0.73	0.80
	SC2	0.61	0.71	0.81
	SC3	0.76	0.82	0.83
	SC4	0.70	0.74	0.74
Value	VS1	0.70	0.71	0.71
	VS2	0.71	0.70	0.59
	VS3	0.78	0.80	0.80
	VS4	0.78	0.80	0.78
	VS5	0.78	0.77	0.80
	VS6	0.77	0.79	0.82
	VS7	0.61	0.65	0.64
	VS8	0.70	0.73	0.77
School	BS1	0.63	0.65	0.69
	BS2	0.59	0.64	0.68
	BS3	0.73	0.79	0.83
	BS4	0.71	0.76	0.75
Teacher	TE1	0.55	0.60	0.66
	TE2	0.74	0.80	0.85
	TE3	0.80	0.84	0.86
	TE4	0.80	0.84	0.88
	TE5	0.77	0.80	0.74
	TE6	0.70	0.74	0.71
	TE7	0.69	0.74	0.72

displays the loading of Saudi Arabia, Bahrain, and Singapore. From Table 6, Saudi Arabia initially included two variables: ”I wish I did not have to study” (Int2) and “Science is boring” (Int3), both of which exhibited weak loadings of 0.37 and 0.48, respectively. These weakly loaded variables will be removed sequentially until the model consists only of variables with loadings greater than 0.5. For Bahrain, the factor of interest contains the variable “I wish I did not have to study” (Int2), with a weak loading of 0.46. Similarly, in Singapore, the factor of interest includes one variable, “I like to conduct science experiments” (Int8), which has a weak loading of 0.49. These weakly loaded variables will be eliminated to achieve a model that satisfies the unidimensionality criterion.

Figure 3, shows the CFA models for Saudi Arabia, Bahrain, and Singapore, respectively. They show that all factor loading values exceed 0.5 after removing low-loading factors. As shown in all figures, the arrows indicate the direction from each factor to its corresponding variables, with the factor loadings displayed. For the Saudi model, the lowest factor loading values within each latent variable are as follows: Interest (Int7 = 0.54), SC (SC2 = 0.61), Value (VS7 = 0.60), School (BS2 = 0.59), and Teacher (TE1 = 0.52). For the Bahrain model, the lowest factor loading values within each latent variable are as follows: Interest (Int3 = 0.52), SC (SC2 = 0.71), Value (VS2, VS7 = 0.67), School (BS2 = 0.64), and Teacher (TE1 = 0.60). For the Singapore model, the lowest factor loading values within each latent variable are as follows: Interest (Int7 = 0.62), SC (SC4 = 0.74), Value (VS2 = 0.59), School (BS2 = 0.68), and Teacher (TE1 = 0.66). These results indicate that every item achieves unidimensionality.

To improve the model, modification indices (MIs) are used. An MIs can be calculated for each fixed parameter, such as factor loadings and error covariances [53]. Notably, modification indices suggest adding relationships, such as covariances between errors, that would enhance the model. For example, the relationship between any two variables, corresponding to the highest modification index value, would be added to the model. If the model still needs improvement, other relationships are added one at a time, and each time, the goodness-of-fit indices and modification indices are recalculated.

For the Saudi model, the correlations are added one by one for the following errors: e7–e8 with a value of 742.49, e16–e17 with a value of 733.34, e14–e15 with a value of 421.14, e20–e21 with a value of 343.46, and e26–e27 with a value of 222.43. For the Bahrain model, correlations are added one by one for the following errors: e20–e21 with a value of 884.68, e18–e19 with a value of 491.89, and e7–e8 with a value of 396.76. For the Singapore model, correlations are added one by one for the following errors: e2–e3 with a value of 1037.26, e20–e21 with a value of 836.75, and e24–e25 with a value of 375.06. Adding correlations between errors with high variance in the modification index has improved the goodness-of-fit of the model.

Figure 3, shows the model after modification.

Table 7. Model Fit Indices after Modified CFA Model.

Name of Category	Name of Index	Level of Acceptance	Saudi Model	Bahrain Model	Singapore Model	Reference
Absolute Fit	RMSEA	$\le 0.07$	0.038	0.048	0.052	[49]
	GFI	$\ge 0.90$	0.957	0.932	0.920
	${\chi }^2$/df	$\le 5.00$	9.220	14.159	14.265
Incremental Fit	CFI	$\ge 0.90$	0.963	0.947	0.943	[49]
	TLI	$\ge 0.90$	0.958	0.941	0.937
	NFI	$\ge 0.90$	0.958	0.943	0.939
Parsimonious Fit	PGFI	$\ge 0.05$	0.803	0.791	0.780	[50]
	PNFI	$\ge 0.05$	0.859	0.854	0.850

Note. RMSEA: Root Mean Square Error; GFI: Goodness Fit Index; CFI: Comparative Fit Index; TLI: Tucker-Lewis fit index; NFI: Normed Fit Index; PGFI: Parsimonious Goodness-of-Fit Index; PNFI: Parsimonious Normed Fit Index.

shows that the model’s fitness indices values after modification have all attained the required level of acceptance. It is observed that the ${\chi }^2$/df value did not reach the acceptable level due to its sensitivity to the sample size as we mentioned previously.

4.3. Validity and Reliability

Once unidimensionality is reached, the model’s validity should be assessed. The construct’s capacity to satisfy each requirement for fit indices is tested for validity. Validity refers to how accurately a concept is measured in the quantitative study. However, reliability is the degree to which a research tool can consistently produce the same results when used in the same circumstances repeatedly. It is the second metric for evaluating the quality of a quantitative investigation [54]. There are two types of validity, namely, convergent and discriminant. There are several scales to measure validity, such as average extracted variance, maximum shared variance, and the square root of average extracted variance [55]. The examination of validity and reliability are presented below.

4.3.1. Reliability

Reliability is defined as the stability of the results when the experiment is conducted repeatedly using a similar approach [56]. The reliability of the measurement model is measured by composite reliability (CR). If the CR value is greater than 0.6, the model is considered reliable [55].

Table 8. Validity and Reliability Indices.

Saudi Model
Factors	CR	AVE	Interest	SC	Value	School	Teacher
Interest	0.898	0.565	0.751
SC	0.792	0.489	0.395	0.699
Value	0.896	0.519	0.716	0.252	0.720
School	0.763	0.448	0.369	0.148	0.327	0.669
Teacher	0.883	0.522	0.664	0.239	0.621	0.350	0.722
Bahrain Model
Factors	CR	AVE	Interest	SC	Value	School	Teacher
Interest	0.908	0.560	0.748
SC	0.838	0.565	0.463	0.751
Value	0.905	0.545	0.708	0.283	0.738
School	0.805	0.511	0.332	0.147	0.346	0.715
Teacher	0.909	0.591	0.670	0.278	0.617	0.391	0.769
Singapore Model
Factors	CR	AVE	Interest	SC	Value	School	Teacher
Interest	0.921	0.599	0.774
SC	0.881	0.649	0.644	0.805
Value	0.902	0.535	0.655	0.366	0.734
School	0.826	0.545	0.299	0.132	0.295	0.738
Teacher	0.913	0.603	0.580	0.306	0.487	0.316	0.777

Note. Bold value: SAVE. Underline value: AVE ≤ 0.5.

presents the reliability test findings for the Saudi, Bahrain, and Singapore models, respectively. In all three models, the lowest Composite Reliability (CR) value is observed in the School variable, 0.763 for Saudi, 0.805 for Bahrain, and 0.826 for Singapore. Each value exceeds the minimum standard of 0.6, indicating that reliability is achieved for every construct in all models.

4.3.2. Convergent Validity

Convergent validity is the measurement of the degree of correlation between many variables of the same construct [57]. If the value of the average variance extracted (AVE) reaches the minimal level (0.5) in each construct in the model, convergent validity is attained.

From Table 8, all AVE values for the Saudi model are greater than 0.5, except for the SC and School constructs, with AVE values of 0.489 and 0.448, respectively. However, an AVE value above 0.4 is acceptable if the construct’s convergent validity is greater than 0.6 [58], which implies that convergent validity is satisfied for the SC and School constructs and all constructs in the Saudi model. For the Bahrain and Singapore models, the lowest AVE values are 0.511 for the School construct and 0.538 for the Value construct, respectively. Since all AVE values in these models exceed 0.5, convergent validity is also satisfied for all constructs in both models.

4.3.3. Discriminant Validity

This is the range of variation between constructs from one another experimentally. It assesses how the interconnected constructs differ from one another [57]. The construct’s elements in the model should not be redundant with one another. The discriminant validity is measured by the correlation between constructs and the square root of the AVE (SAVE). The discriminant validity is fulfilled when the SAVE value for each construct has a greater value than the correlation values for this construct and all other constructs.

From Table 8, the value of the SAVE values is represented by the bold number, while the values below these numbers represent the correlation between the constructs. For Saudi model, the SAVE values are: Interest = 0.751, SC = 0.699, Value = 0.720, School = 0.669, and Teacher = 0.722.For Bahrain model, the SAVE values are: Interest = 0.748, SC = 0.751, Value = 0.738, School = 0.715, and Teacher = 0.769. For Singapore model, the SAVE values are: Interest = 0.774, SC = 0.805, Value = 0.734, School = 0.738, and Teacher = 0.777. Each of these SAVE values is greater than the corresponding construct correlation values, confirming that discriminant validity has been achieved for all models.

4.4. Structural Equation Modeling

Following conducting the CFA in Section 4.2, the causal relationships between the constructs are built to implement the SEM models. Bootstrap based on the maximum likelihood approach is performed in the SEM models to produce unbiased estimates, with 1000 bootstrap samples.

In the aforementioned models, the latent variables are Self-Concept (SC), Teacher Effectiveness (Teacher), Belonging to School (School), Interest, and Value of Science (Value). The response variable is Future Job (Job). Notably, SC, Teacher, and School are exogenous variables, so they are independent variables. Job is an endogenous variable, so it is a dependent variable. Interest and Value are endogenous and exogenous variables at the same time, and they are mediator variables (See Figure 4). Figure 4 shows the SEM models for Saudi Arabia, Bahrain, and Singapore, respectively.

4.5. Mediation Analysis

The present study has two potential factors previously identified as mediating factors, which are the value of science and interest in science [11,14]. The aim of the mediation analysis employed in this study is to investigate whether value and interest in science mediate the influence of the independent variables (self-concept, teacher, and school) on the response variable (future job) (see Figure 4). The direct effects of the independent variables on the response variable are calculated, as well as the indirect effects in the presence of the mediators. Additionally, the total effects are calculated, which reflect the total of the independent variables’ direct and indirect effects.

Table 9. Mediation Analysis for Saudi Arabia, Bahrain, and Singapore.

Path	Saudi Estimation	Bahrain Estimation	Singapore Estimation
Path from predictor to mediator1
SC → Interest	0.25 *	0.30 *	0.52 *
School → Interest	0.14 *	0.07 *	0.11 *
Teacher → Interest	0.59 *	0.58 *	0.39 *
Path from predictor to mediator2
SC → Value	0.12 *	0.14 *	0.27 *
School → Value	0.12 *	0.11 *	0.15 *
Teacher → Value	0.58 *	0.56 *	0.37 *
Path from mediator to response
Interest → Job	0.27 *	0.28 *	0.33 *
Value → Job	0.68 *	0.72 *	0.61 *
Indirect Effect
SC → Interest → Job	0.07 *	0.08 *	0.17 *
SC → Value → Job	0.08 *	0.10 *	0.16 *
School → Interest → Job	0.04 *	0.02 *	0.03 *
School → Value → Job	0.08 *	0.08 *	0.09 *
Teacher → Interest → Job	0.16 *	0.16 *	0.12 *
Teacher → Value → Job	0.39 *	0.40 *	0.22 *
Total Indirect Effect
SC → Interest → Job + SC → Value → Job	0.15 *	0.18 *	0.33 *
School → Interest → Job + School → Value → Job	0.12 *	0.10 *	0.12 *
Teacher → Interest → Job + Teacher → Value → Job	0.55 *	0.56 *	0.34 *
Direct Effect
SC → Job	−0.07 *	−0.06 *	0.02 *
School → Job	−0.03 *	−0.04 *	−0.05 *
Teacher → Job	−0.16 *	−0.22 *	−0.13 *
Total Effect
SC → Interest → Job + SC → Value → Job + SC → Job	0.08 *	0.12 *	0.35 *
School → Interest → Job + School → Value → Job + School → Job	0.09 *	0.06 *	0.07 *
Teacher → Interest → Job + Teacher → Value → Job + Teacher → Job	0.39 *	0.34 *	0.21 *

* p-value: * ≤ 0.05.

presents the direct, indirect, and total effects of all three countries. The numbers in the table refer to the path regression coefficients in standardized form. These coefficients are measured on a scale from −1 to 1 and show the strength of the relationship between two variables. The following sections provide a detailed analysis of the factors effect.

4.5.1. Mediation Analysis for Saudi Arabia

Table 9 presents the direct and indirect effects of the predictor variables (SC, Teacher, and School) on the Job variable for Saudi students. From the table, both mediators, Interest and Value, significantly affect the Job, with Value having a higher impact (0.68) than Interest (0.27). There is a direct effect of the independent variables (SC, School, Teacher) on the dependent variable (Job), but it is relatively weak, while the indirect effect is larger and significant.

The indirect effect is obtained by multiplying the path value between the independent variable and the mediator by the path value between the mediator and the response variable. The total effect is the sum of direct and indirect effects.

Furthermore, Table 9 indicates that all predictors significantly influence the mediators, Interest and Value. The Teacher construct has the highest effect, with values of 0.59 and 0.58 on Interest and Value, respectively. In contrast, the SC and School variables have an impact but less impact than the Teacher construct. The indirect effect is significant for all variables. The Teacher variable has the highest indirect effect on the Job through the Interest and Value mediators, with values of 0.16 and 0.39, respectively. Thus, the teacher has the highest total indirect effect with a value of 0.55. On the contrary, the School variable is found to be the least influential in all cases, with a total effect of 0.09. These results indicate that Interest and Value partially mediate the relationship between self-concept, teacher effectiveness, and school belonging with the response variable (Future Job). Therefore, self-concept, teacher effectiveness, and school belonging directly influence the choice of a future job and indirectly through Interest and Value in Saudi students.

4.5.2. Mediation Analysis for Bahrain

Table 9 shows the direct and indirect effects of the predictor variables (SC, Teacher, and School) on the future job for Bahraini students. The mediators, Interest, and Value, significantly impact the future job, with Value having a stronger effect (0.72) compared to Interest (0.28). All independent variables exhibit a weak direct effect on the response variable (Future Job).

According to Table 9, all predictors significantly influence the mediators, Interest and Value. The Teacher construct has the highest impact, with effects of 0.58 on Interest and 0.56 on Value. In contrast, the School construct shows the least effect on the mediators, with 0.07 on Interest and 0.11 on Value. All predictors have a significant indirect effect through the mediators. The Teacher variable provides the highest indirect effect with a value of 0.56. For the total effect, the teacher variable also has the highest total effect (0.34). Conversely, the School variable is observed to be the least influential across all cases, and its total effect was 0.06. Although it is significant, it is weak. These results indicate that Interest and Value partially mediate the relationship between self-concept, teacher effectiveness, and school belonging with the response variable (future job). This means that self-concept, teacher effectiveness, and school belonging directly influence future job choice and indirectly influence it through Interest and Value.

4.5.3. Mediation Analysis for Singapore

Table 9 outlines the direct and indirect effects of the predictor variables (SC, Teacher, and School) on future job outcomes in Singapore. From the table, it is evident that both mediators, Interest and Value, significantly impact the future job, with Value exerting a stronger influence (0.61) than Interest (0.33). All predictor variables have a significant direct effect on the mediator variables, Interest, and Value. The SC predictor has the highest impact on Interest, with a value of 0.52, and the Teacher predictor has the highest impact on Value, with a value of 0.37. However, the School construct has a weak impact on the mediators compared to the other variables. The direct effect of the predictors on future job is weak.

According to the indirect effect, the SC variable provides the highest indirect effect with a value of 0.17 through Interest, while the Teacher variable provides the highest indirect effect with a value of 0.22 through Value. For the total effect, the SC variable also has the highest total effect, with a value of 0.35. However, the School variable provides a weak total effect with a value of 0.07. These results indicate that the Interest and Value partially mediate the relationship between the predictors (self-concept, teacher effectiveness, and school belonging) and the response variable (future job) in Singapore. In other words, self-concept, teacher effectiveness, and school belonging have both direct and indirect influences on future job choice through Interest and Value.

5. Summary

The mediation analyses for Saudi Arabia, Bahrain, and Singapore reveal both consistent trends and notable differences in how various factors influence students’ future job choices. Across the three countries, interest in science and the perceived value of science significantly shape students’ career aspirations in STEM fields. In particular, the value of science exerts a stronger influence on career decisions than the interest in all models, suggesting that students are more likely to pursue science related careers when they perceive its long-term benefits and relevance to their future success. In addition, the results indicate that the impact of value and interest in science on career choices is greater than the direct effects of self-concept, school belonging, and teacher effectiveness. This implies that while these factors may not have a strong direct effect on career aspirations, they indirectly influence career choices by shaping students’ interest in and appreciation of science.

The role of teacher effectiveness, self-concept, and school belonging differs across the three countries:

• For Saudi Arabia and Bahrain:

• Teacher effectiveness has the strongest overall influence on students’ career choices, with total effects of 0.39 and 0.34, respectively.
• Self-concept (students’ confidence in their scientific abilities) plays a weaker role, with total effects of 0.08 (Saudi Arabia) and 0.12 (Bahrain).
• School belonging has a relatively small impact in both countries, with total effects of 0.09 (Saudi Arabia) and 0.12 (Bahrain).

These findings suggest that teacher engagement and effectiveness play a crucial role in fostering students’ enthusiasm for science in Saudi Arabia and Bahrain. Investing in professional development programs for teachers could significantly enhance students’ motivation and long-term interest in STEM careers.

• For Singapore:

• Self-concept has the strongest influence on career choices, with a total effect of 0.35, indicating that students’ confidence in their science abilities is the key driver of STEM aspirations.
• Teacher effectiveness follows with a total effect of 0.21, suggesting that while teachers still play a role, students’ personal confidence is more decisive in shaping their career expectations.
• School belonging has the weakest influence, with a total effect of 0.07, implying that school climate and peer relationships play a relatively minor role in STEM career aspirations in Singapore.

These results suggest that Singapore’s educational system fosters a strong sense of self-efficacy in students, which may contribute to their higher motivation and independence in pursuing STEM careers.

This difference in results may be influenced by the unique emphasis in Singapore’s educational system on fostering self-concept and confidence. The system strongly promotes self-directed learning, encouraging students to develop critical thinking skills, resilience, and effective communication. By nurturing curiosity, flexibility, and persistence, Singapore’s education aims to produce self-assured, self-reliant thinkers. This focus on developing students into lifelong learners who take ownership of their development may help explain why the self-concept factor has a larger influence on Interest, while the teacher effectiveness factor remains crucial for shaping Value in the Singapore model [59].

To further enhance self-directed learning, Singapore has adopted blended learning, which combines in-class and home-based instruction. This model, accelerated by the COVID-19 pandemic, encourages students to manage their education through scheduled home based learning days. Flexible learning frameworks and personal learning devices allow students to study courses at their own speed while still making academic progress in a planned manner. This method gives pupils self-confidence, freedom, and motivation, in addition to encouraging digital literacy [59,60].

6. Conclusions

This study explores the factors influencing students’ job choices in science, examining both the direct and indirect effects of these factors on future careers through structural equation modeling. It compares the influences on students in Saudi Arabia, Bahrain, and Singapore.

The results reveal that interest in and value of science among eighth-grade students in Saudi Arabia, Singapore, and Bahrain strongly influence their decisions to pursue science-related careers. Moreover, the indirect effects of self-concept, school belonging, and teacher influence on career choice—mediated by values and interests—are both strong and significant. This indicates that values and interests play a positive role in these mediation relationships.

The findings highlight the significant role of teachers in enhancing the perceived importance of science subjects and cultivating students’ passion for them, ultimately impacting their career choices. Additionally, self-concept emerges as a critical factor, significantly affecting students’ interest in science and, consequently, their career decisions.

This suggests that educational interventions should prioritize enhancing students’ interest in and value of science, along with factors that can strengthen these, such as teacher engagement and self-concept development.

To console the self-concept, Saudi Arabia could adopt Singapore’s blended learning model, which combines in-class instruction with home-based learning. It can implement blended learning gradually by starting with pilot schools before nationwide expansion. Government needs to invest in digital infrastructure, ensuring internet access and personal learning devices for all students. Teacher training is essential to equip educators with skills for blended and online learning. A flexible curriculum should incorporate student-initiated learning to foster curiosity beyond traditional subjects. Lastly, parental and community support is crucial in sustaining self-directed learning. By adopting these strategies, independent, confident, and future-ready learners could be developed [59,60].

Given the influential role of the teacher, supporting teachers’ professional development is essential to maximize their positive effects on student outcomes. To strengthen teacher development in STEM education, we recommend implementing several strategies, including mentorship programs that pair experienced STEM educators with newer teachers and foster industry partnerships for real-world insights. Additionally, offering professional development workshops focused on STEM, such as inquiry based learning, project based learning, and the integration of technology in the classroom, will equip teachers with effective teaching methods. Encouraging cross-disciplinary collaboration among STEM teachers and providing training on the latest educational technologies can further enhance teaching practices. Finally, incorporating culturally relevant STEM content tailored to the local context will ensure that the curriculum is engaging and accessible for diverse student populations. These strategies will support the professional growth of teachers and improve the quality of STEM education [61,62,63,64,65].

These findings are consistent with multiple studies that underscore the necessity of enhancing teachers’ competencies in STEM education as a critical measure toward achieving sustainable development goals. Furthermore, the perceptions of teachers regarding STEM education significantly influence the effectiveness of their instructional practices. Educators who maintain a positive outlook and possess a comprehensive understanding of STEM are more effectively positioned to guide students toward careers in science and technology. This, in turn, cultivates interest in these fields and promotes meaningful contributions to sustainable development [19].

On the other hand, to enhance STEM education and better prepare students for future careers, curriculum development should focus on aligning educational content with industry demands, ensuring students gain relevant and practical skills. A competency based approach should replace traditional content-heavy instruction, emphasizing critical thinking, problem-solving, and innovation. Integrating STEM disciplines through interdisciplinary learning can enhance students’ ability to apply knowledge in real-world contexts. Additionally, incorporating emerging technologies such as artificial intelligence, data science, and robotics will ensure that curricula remain relevant in a rapidly evolving digital landscape. To support diverse learning pathways, flexible and modular curriculum structures should be introduced, allowing students to explore specialized STEM fields based on their interests and career aspirations. Implementing these curriculum enhancements will strengthen STEM education, fostering a workforce capable of driving innovation and economic growth [66].

7. Research Limitations

The research examines several factors that may influence students’ attitudes toward STEM careers. However, due to this specific focus, the study may not fully capture all elements that could play a significant role in shaping students’ attitudes. The school, home, and community environments encompass numerous interconnected factors that could impact students’ decisions, making it challenging to represent them all within a single model. Additionally, the available data may not encompass all relevant factors.