Understanding the influence of business innovation context on intentions of enrolment in master education of STEM students: a multi-level choice model

The study investigates why final-year bachelor students pursue master’s degrees, with emphasis on STEM versus non-STEM differences. Building on human capital theory and cultural/social capital frameworks, it examines how individual characteristics, university context, and local economic conditions shape the predisposition to continue to master’s study immediately after graduation. Prior work shows heterogeneity in graduate enrolment by demographics, academic performance, parental education and income, institutional quality/type, and labour market expectations. The paper argues that expected returns and local labour market conditions, including innovation intensity, can differentially affect STEM and non-STEM students. The research aims to propose and test a two-level model incorporating individual/organizational habitus (level 1) and local business innovation environment (level 2). Research questions: RQ1: Do STEM vs. non-STEM majors have different effects on intentions to pursue master’s education across local contexts? RQ2: How does the business innovation context interplay with field of study in shaping predisposition to enrol in a master’s program?

The literature links educational progression to a combination of individual, institutional, and economic factors. Individual-level determinants include gender, age, residence, race/ethnicity, academic performance, parental education (cultural capital), socioeconomic status, educational debt, and labour market participation. Institutional influences encompass university quality, type (research vs. teaching-oriented), and field of study, with science fields often showing higher continuation rates than arts/humanities. From human capital theory, expected returns (wages, employment probabilities), unemployment rates, and broader economic context affect demand for further education, with higher expected returns encouraging postgraduate enrolment. Technological change and innovation raise demand for skills, potentially altering returns to further study and varying across regions. Prior Romanian evidence suggests that regions with higher employment in science/technology raise master enrolment while more dynamic innovation and earnings contexts may discourage prolonging education due to higher forgone earnings. These strands motivate examining cross-level interactions between field of study and local business innovation.

Design: Multi-level (hierarchical) modeling of intention to enrol in a master’s program in the next academic year among final-year bachelor students. Data: Cross-sectional survey in 2019 of 502 students (age M=22.08, SD=1.185; 54.4% men) from 10 Romanian public universities across regions; fields coded into STEM (n=250) and non-STEM (n=252). Measures (Level 1): demographics (gender, age, urban/rural residence), academic performance (average grade), parental education (father’s education), subjective family income, employment status (not working/part-time/full-time), work seniority, expectations about master-graduate wages and unemployment; university performance score (from national university ranking, 2019). Measures (Level 2): regional business innovation indicators from official statistics—shares of enterprises introducing product innovations, process innovations, and both. Outcome: Binary indicator of intention to enrol in a master’s program the following year (2020/2021). Modeling strategy: Multilevel mixed-effects logistic regression. Steps: (1) Null two-level random intercept model to test between-region variability; (2) Random intercept model with individual-level variables; (3) Add regional-level innovation indicators; (4) Test cross-level interactions between field of study (STEM vs non-STEM) and regional innovation; (5) Random slope models allowing the STEM coefficient to vary across regions; likelihood ratio tests to compare models; report odds ratios and variance components (variance partition coefficients at region and university levels). Estimation: maximum likelihood with adaptive quadrature; tests include Wald and LR tests.

- Descriptives: 53.6% of students intend to enrol in a master’s program next year; STEM 62% vs non-STEM 45.2%. Significant differences by region/university (Kruskal–Wallis), with Bucharest-Ilfov highest. - Variance components: Region-level null model shows between-region variance ~0.094 with VPC ≈ 2.77% of residual variation attributable to regions. Between-university variance ~0.81, VPC ≈ 19.75% (also ~21.44% and ~18.28% in specific models). - Individual-level effects (robust across models): Higher grades increase odds; father’s higher education increases odds; full-time employment increases odds; longer work seniority decreases odds. Gender, age, residence area, subjective income, and perceived master-level wage/unemployment largely not significant. - Field of study: STEM students have higher propensity to enrol than non-STEM. In multilevel models, the non-STEM indicator is negative and significant (e.g., exp(β) ≈ 0.50). - University performance: Higher university score positively associated with intention (e.g., OR ≈ 1.05–1.07 per unit). - Regional innovation: Higher share of enterprises introducing product innovations reduces intention to pursue a master’s (e.g., β ≈ −0.52*, OR ≈ 0.59; up to β ≈ −1.07***, OR ≈ 0.34 depending on model). Process-only or combined product+process indicators show no significant effects. - Cross-level interaction: The negative effect of product innovation share is stronger for STEM students. Interaction term positive and significant (e.g., β ≈ 1.00***, OR ≈ 2.72; in random-slope model β ≈ 1.06**, OR ≈ 2.90), indicating that as product innovation increases, STEM students are more deterred from enrolling than non-STEM students. - Random slope results: Significant regional variation in the STEM vs non-STEM gap (LR tests significant). Intercept–slope covariance negative, implying regions with higher overall master-enrolment propensity tend to have smaller STEM effects. Between-region variation larger for STEM (variance ≈ 1.137) than for non-STEM (variance ≈ 0.062).

Findings align with human capital theory and cultural capital: academic achievement and parental (father’s) higher education increase intentions to pursue master’s study; university performance also matters, suggesting institutional quality/fit influences continuation. Labour market engagement has nuanced effects: full-time workers are more inclined to enrol—possibly seeking specialized skills that complement work—while longer work seniority reduces demand for master’s study, implying substitution by experience. Regional context matters: innovative business environments appear to pull graduates into immediate employment, lowering the propensity to continue to master’s education. This effect is more pronounced for STEM students, whose skills are in higher immediate demand in innovative firms, raising forgone earnings. Non-STEM students’ intentions are less sensitive to regional innovation intensity, suggesting more uniform returns across regions. The documented regional heterogeneity in the STEM vs non-STEM gap underscores that local economic structures shape educational decisions in conjunction with individual characteristics.

The paper proposes and validates a two-level choice model for master’s enrolment that integrates individual/organizational habitus with regional business innovation conditions. Empirically, STEM students are generally more likely to intend enrolling in master’s programs, but this advantage varies across regions and diminishes in more innovative business environments that incentivize immediate labour market entry. Academic performance, parental education, full-time work, and university quality increase intentions, while longer work seniority reduces them. Policy implications include strengthening university–industry linkages, providing flexible/dual master’s pathways, and financial supports that mitigate opportunity costs, particularly in highly innovative regions. Future research should extend to longitudinal designs tracking outcomes, cross-country comparisons across different innovation ecosystems, and mixed-methods approaches to capture motivations and contextual nuances.

Generalizability is limited by the study’s focus on Romanian public universities and specific regional innovation contexts. Cross-sectional design with intentions (not actual enrolment) constrains causal inference. Indicators of innovation are regional aggregates that may not capture firm-level heterogeneity or specific industry linkages. Some expected-return measures (perceived wages/unemployment) showed no effects, possibly due to measurement or timing. Broader coverage of fields (e.g., medical, sports, military/defence excluded), institutions, and countries, along with longitudinal data, would strengthen external validity.