An Analysis of Factors Influencing SW·AI Career Goals of Students Participating in the Youth SW Companion Project

2.1 Characteristics and Significance of the Youth SW Companion Project

The Youth SW Companion Project is an initiative launched in 2024 by the Ministry of Science and ICT and KOSAC, aimed at strengthening middle and high school students' digital-based self-directed problem-solving competencies and enhancing their career adaptability in the SW·AI field [14]. The initiative was designed to complement school-based informatics education in line with the expanded informatics instruction hours under the 2022 Revised National Curriculum, and to connect school-based learning with external digital education resources. Jeon, S. et al. (2023) [2] analyzed how the AI education content framework in the 2022 Revised Curriculum was structured to systematically link instruction across all grade levels from elementary through high school, and recommended an approach that combines AI education opportunities both inside and outside of school.

The program's operational structure is characterized by an ecosystem-based model in which four stakeholders collaborate: student-led clubs, supporting teachers, university student mentors providing tutoring and career counseling, and professional instructors from digital innovation companies offering career exploration opportunities. Educational content is organized around digital problem-solving programs linked to ten topics selected from the UN SDGs' 17 goals based on youth interests including marine and terrestrial ecosystem conservation, climate change response, and reduction of inequalities and is delivered in a blended online-offline format totaling 15 hours. In 2024, 4,360 students nationwide and 1,241 university student mentors participated [14].

Instruction was delivered through Project-Based Learning (PBL). Rather than passively receiving knowledge, students were guided to internalize competencies through the process of independently planning, implementing, and presenting digital projects connected to real-world social problems. The program was conducted in a blended (online and offline) format over 15 sessions, following a sequence of: exploration of social problems and selection of an SDGs topic, data collection and exploration of digital tools, prototype design and implementation using coding or data analysis tools, and public presentation of outcomes.

In this way, the Youth SW Companion Project is distinguished from conventional SW education programs in that it provides experiences in which students implement tangible outcomes using digital technologies within the context of SDGs-linked real-world social problems, and incorporates informal career modeling through university student mentors. This is consistent with research findings emphasizing that experiences in which students proactively identify and solve problems are more effective than knowledge-transmission approaches in fostering digital competencies and career motivation [3].

2.2 Career Goals in SW·AI and Self-Efficacy

Career aspirations in the SW·AI field are growing rapidly among Korean youth. According to the Ministry of Education's (2023) Survey on Career Education in Elementary and Secondary Schools, the proportion of high school students aspiring to careers in emerging technology sectors such as software development, IT engineering, and robotics engineering increased approximately threefold from 3.6% in 2013 to 11.6% in 2023, while the proportion among middle school students rose from 3.5% to 5.3% over the same period [15]. This can be understood as a reflection of adolescents' keen sensitivity to the societal changes brought about by accelerating digital transformation. However, a survey of AI perceptions conducted by Kwon, D., Heo, N., & Kang, J. (2023) [16] among 272 high school students found that while students scored high on perceiving AI as necessary in their lives, their interest in learning it was comparatively low. This suggests a gap between awareness of AI's necessity and motivation for AI-related career orientation, and indicates the need for educational approaches that go beyond simply raising AI awareness to internalize career motivation.

Self-efficacy is attracting attention as a key mechanism for bridging this gap. Self-efficacy refers to an individual's belief in their ability to successfully perform a specific task; according to social cognitive career theory, it directly influences the formation of career interests and the setting of career goals [17, 18]. In the context of SW·AI education, Oh, G., & Jang, E.(2021) [8] empirically demonstrated that university students' SW education experience significantly affects AI confidence and computational thinking self-directedness, while Park, C., & Kang, G. (2022) [19] confirmed that self-efficacy serves as a core antecedent variable of AI service acceptance attitudes, exerting a positive influence through the mediation of innovativeness. Park, S. (2024) [20] also reported that participation in extracurricular programs has the potential to enhance self-efficacy among late adolescents, suggesting that program experiences conducted outside of school such as the SW Companion Project can contribute to self-efficacy development. These studies consistently support the view that competency self-efficacy in the SW·AI domain is a core factor promoting technology adoption and career-oriented behavior.

Nevertheless, studies directly examining the effect of SW·AI education on students' career goals remain scarce. Lee, Y. (2024) [13], in a meta-analysis of 191 papers on informatics education, noted that studies pertaining to career attitudes and motivation account for only a small fraction of the total, with the majority focused on improvements in cognitive competencies. Sim, T., & Ye, C. (2023) [12], in their analysis of domestic SW·AI education research trends, similarly highlighted the pronounced shortage of research on career motivation and self-efficacy among affective competencies relative to cognitive competency research. Furthermore, studies that simultaneously examine teacher-related and student-related factors in an integrated manner to assess their influence on SW·AI career goals are virtually nonexistent.

This study therefore seeks to fill this gap by integrating analysis of students' competency self-efficacy, SDGs awareness, program satisfaction and quality perceptions, and teachers' program delivery perceptions to empirically identify the factors that influence SW·AI career goals.

3.1 Participants and Data Collection

To analyze the factors influencing career aspirations in SW·AI, this study utilized survey data collected through the "2024 Youth SW Companion Project” conducted by KOSAC. The survey was administered between July and November 2024 to 163 teachers (36 middle school, 127 high school) and 1,090 students (282 middle school, 808 high school) who participated in the project across all 17 metropolitan cities and provinces nationwide. Prior to data collection, participants were informed that the collected data may be used for research purposes in accordance with confidentiality protections under Article 33 of the Statistics Act. Prior to analysis, cases with missing values on key variables were excluded, and only complete-response data were used in the final analysis. Table 1 presents the number of participating teachers and students across all 17 metropolitan cities and provinces.

Table 1.

Number of Participating Teachers and Students by Region

3.2 Measurement Instruments

The survey instruments were divided into a student version and a teacher version. The student survey comprised four constructs SDGs Awareness (SA1), Digital Problem-Solving Competency Self-Efficacy (SA2), Program Satisfaction (SB1), and Perceived Program Quality (SB3) along with the dependent variable, Career Goals in the Digital (SW·AI) and Computer Fields (SB2). The teacher survey consisted of four constructs: Program Design (T1), Instructional Content (T2), Instructional Methods (T3), and Overall Evaluation (T4). All items were measured on a five-point Likert scale (1 = Strongly Disagree to 5 = Strongly Agree). Table 2 presents the composition of the measurement instruments. The survey was based on the "2024 Youth SW Companion Project Effectiveness Survey" conducted by KOSAC through Gachi Solutions in December 2024, and the detailed survey items are provided in the Appendix.

Table 2.

Measurement Scales and Constructs

3.3 Research Model

The theoretical foundation of this study is Bandura's (1997) [17] theory of self-efficacy and Lent, Brown, & Hackett's (1994) [18] Social Cognitive Career Theory (SCCT). According to SCCT, an individual's career goals are shaped by self-efficacy and outcome expectations, which develop through the mediation of learning experiences. This study applies this theoretical framework to the context of informal SW·AI learning among adolescents and posits three pathways as follows.

The first pathway is competency self-efficacy. Digital problem-solving competency self-efficacy (SA2) serves as a core antecedent variable in SCCT; it is hypothesized that students' belief in their ability to solve problems using digital technologies exerts a direct influence on the formation of SW·AI career goals.

The second pathway is outcome expectations and social value awareness. SDGs awareness (SA1) corresponds to the outcome expectations mechanism in SCCT; it is hypothesized that students' perception that digital technology can contribute to solving social problems exerts a positive influence on the formation of career consciousness.

The third pathway is the quality of the learning environment. Program satisfaction (SB1), perceived program quality (SB3), and teacher-related variables (T1–T4) are hypothesized to function as learning environment conditions within which self-efficacy and outcome expectations are formed, thereby exerting an indirect influence on career goals.

3.4 Analytical Methods

The dependent variable was students' "Career Goals in the Digital (SW·AI) and Computer Fields" (SB2), computed as the mean of the five items comprising that construct. Independent variables were computed as mean factor scores for the items within each construct, consisting of four student variables (SA1, SA2, SB1, SB3) and four teacher variables (T1, T2, T3, T4).

To achieve the research objectives, Multiple Regression Analysis was conducted using SPSS 26.0. Model 1 included student variables only, and Model 2 additionally incorporated teacher variables. VIF values were computed for all predictors to assess multicollinearity, and standardized residuals were examined to confirm the absence of influential outliers.

However, multiple regression analysis assumes linearity and independence among variables and does not directly decompose the contribution of each predictor to individual predicted values. To address these limitations, SHAP (SHapley Additive exPlanations) analysis was additionally performed using Python 3.12. SHAP rigorously decomposes each independent variable's contribution to individual predicted values based on Shapley values from cooperative game theory [21], enabling visual exploration of nonlinear effects and interaction patterns among variables. Its model-agnostic property further allows cross-validation of the convergence between regression and machine learning model results. SHAP analysis was conducted using the SHAP library in Python; a LinearRegression model (scikit-learn) was fitted on the same predictor set as the multiple regression models, and LinearExplainer was applied to compute individual feature contribution values for each observation. The LinearRegression model was applied using scikit-learn's default settings (fit_intercept=True) without additional hyperparameter tuning. This enabled intuitive visualization and complementary interpretation of the rankings and directions of variable influence identified through regression analysis.

4.1 Reliability and Descriptive Statistics of Measurement Instruments

To verify the internal consistency of the measurement instruments, Cronbach's α coefficients were computed for each construct, and descriptive statistics were calculated to characterize the distribution of key variables. Reliability and descriptive statistics are presented in Table 3.

Table 3.

Reliability and Descriptive Statistics of the Measures

The results showed that the dependent variable, SW·AI Career Goals (SB2), had a reliability of α = .871. Among the student independent variables, all demonstrated satisfactory reliability: SA1 SDGs Awareness (α = .854), SA2 Digital Problem-Solving Competency (α = .916), SB1 Program Satisfaction (α = .903), and SB3 Perceived Program Quality (α = .921). The teacher variables likewise demonstrated high reliability across all constructs, with T1 Program Design (α = .871), T2 Instructional Content (α = .929), T3 Instructional Methods (α = .931), and T4 Overall Evaluation (α = .931), all reaching α ≥ .87.

4.2 Multiple Regression Analysis Results for SW·AI Career Goals

To identify the factors influencing SW·AI Career Goals (SB2), a two-step multiple regression analysis was conducted: Model 1 (n = 1,090), which included only student variables, and Model 2 (n = 974), which additionally incorporated teacher variables. Model 2 was restricted to students attending the 107 schools from which teacher survey data were collected, reducing the sample to n = 974.

4.2.1 Model 1: Effects of Student Variables

Table 4 presents the multiple regression results for student variables predicting SW·AI Career Goals. Model 1, which included four student independent variables, was statistically significant (F = 369.35, p < .001) and accounted for 57.7% of the variance in the dependent variable. All four variables exerted significant positive effects on SW·AI Career Goals. (SA2) Digital Problem-Solving Competency (β = .368, t = 11.534) showed the strongest positive effect, indicating that higher student self-efficacy in solving problems using digital technologies is associated with stronger career aspirations in the SW·AI field. (SA1) SDGs Awareness (β = .187, t = 6.074) showed the second strongest significant positive effect, confirming that students with a stronger sense of commitment to addressing sustainable development issues through digital technology report higher career goals. (SB3) Perceived Program Quality (β = .149, t = 3.817) and (SB1) Program Satisfaction (β = .147, t = 3.789) also each exerted significant positive effects.

Table 4.

Results of Multiple Regression Analysis on SW·AI Career Goals (Model 1: Student Variables)

4.2.2 Model 2: Effects of Student and Teacher Variables

Model 2, which additionally incorporated teacher variables, was statistically significant (F = 170.70, p < .001) and explained 58.6% of the variance in the dependent variable (R²adj = .583). The incremental variance explained by the addition of teacher variables over Model 1 (ΔR²= .009) was statistically significant but modest in magnitude, indicating that teacher variables as a set contribute only marginally beyond student variables in explaining SW·AI career goals.

The influence of student variables remained largely consistent with Model 1. Among the teacher variables, T3 Instructional Methods (β = −.170, t = −3.592) showed a statistically significant negative coefficient, and T4 Overall Evaluation (β = .087, t = 2.003) showed a positive coefficient, while T1 Program Design (β = .058, t = 1.741) and T2 Operational Content (β = .019, t = 0.492) were not statistically significant.

The negative coefficient of T3 warrants cautious interpretation. Prior to interpreting the direction of this effect, it should be noted that the four teacher variables showed moderate intercorrelations; however, VIF values for all teacher predictors were below 3.0 (T1 = 2.31, T2 = 2.78, T3 = 2.64, T4 = 2.19), indicating that multicollinearity does not substantially distort the estimates. Nevertheless, given that students are nested within classrooms and schools, a multilevel modeling approach would be more appropriate for isolating the teacher-level effects from within-classroom variance. The present analysis should therefore be understood as an exploratory finding, and the negative association between T3 and student career goals — while statistically significant — should not be interpreted causally without further investigation. One plausible interpretation is that instructional approaches rated highly by teachers may have been perceived by students as overly demanding or misaligned with their current proficiency levels, potentially suppressing career motivation; however, this remains speculative given the cross-sectional design and the absence of student-perceived difficulty measures. Table 5 presents the multiple regression results for student and teacher variables predicting SW·AI Career Goals.

Table 5.

Results of Multiple Regression Analysis on SW·AI Career Goals (Model 2: Student+Teacher Variables)

4.3 SHAP Analysis Results for Variable Contribution

SHAP analysis was conducted to complement the multiple regression findings by visually confirming the distribution of individual variable contributions and providing a more nuanced comparison of relative contribution magnitudes that regression coefficients alone cannot fully capture. The analysis was performed primarily based on Model 1 with student variables, and was additionally conducted for Model 2 including both student and teacher variables.

4.3.1 Model 1: SHAP Analysis Results for Student Variables

The SHAP analysis results showed that the variable importance rankings were fully consistent with the regression coefficient (β) rankings. The SHAP Feature Importance plot in Figure 1 represents the mean |SHAP value| of each variable as bar length, with (SA2) Digital Problem-Solving Competency (SHAP = .2955) showing the highest contribution, followed by (SA1) SDGs Awareness (.1508), (SB3) Perceived Program Quality (.1219), and (SB1) Program Satisfaction (.1179), in that order. Notably, (SA2) Digital Problem-Solving Competency d e m o n s t ra t e d a n o ve r wh el m i n gly h i gh e r contribution magnitude compared to the other three variables, visually confirming that digital problem-solving competency self-efficacy plays a decisive role in predicting SW·AI career goals.

Figure 1.

SHAP Feature Importance for Student Variables (Model 1)

4.3.2 Model 2: SHAP Analysis Results for Student and Teacher Variables

In the SHAP analysis for Model 2 incorporating both student and teacher variables, (SA2) Digital Problem-Solving Competency (SHAP ≈ .30) exhibited overwhelmingly high importance. Among the teacher variables, (T3) Instructional Methods recorded the second highest SHAP importance. This is consistent with the regression finding that T3 exerted a significant negative effect (β = −.170), and the SHAP analysis similarly showed that T3's contribution was relatively larger than that of other teacher variables, confirming that the two analytical approaches converge in a mutually complementary manner.

Figure 2.

SHAP Feature Importance for Student and Teacher Variables (Model 2)