Introduction

“What kind of job do you expect to have when you are about 30 years old?”. This all-time classic question was answered by 15-year-old students who participated in the Organisation for Economic Co-operation and Development’s (OECD) Programme for International Student Assessment (PISA) in 2015 around the world. Focusing on Greek students’ responses to this question, along with data about their demographic background, self-beliefs, attitudes, motivation, and their academic achievement, we examined their expectations to follow professions within the science, technology, engineering, and mathematics (STEM) fields.

Since its introduction in 2001, the STEM acronymFootnote 1 has received multiple interpretations depending on purpose, audience, and context (Lyons, 2020). Although the STEM acronym has been introduced as an effort to integrate some or all of the four disciplines into one (Moore & Smith, 2014), the interdisciplinarity of the term has not yet allowed for a consensus on the specific professions that fall within it (Akerson et al., 2018; Blackley & Howell, 2015; Donahoe, 2013; Lyons, 2020). Consequently, researchers tend to adopt either a wider or a narrower approach to what can be defined as STEM (Marginson et al., 2013). One of the main differences between these two approaches is that health professions (e.g., medical doctors, nurses) are considered STEM based on a wider approach (OECD, 2017b), but non-STEM based on a narrower approach (Hall & Rathbun, 2020; Kaufmann & Wittmann, 2018; Sikora & Pokropek, 2012; Tornese & Lupiañez-Villanueva, 2017).

Wider approaches tend to include not only science, technology, engineering, and mathematics professions but also those related to information and communications technology (e.g., Stoeger et al., 2016), earth sciences (e.g., Nazareth et al., 2019), and health sciences (e.g., Wallace et al., 2015).Footnote 2 Within this framework, the International Standard Classification of Occupations (ISCO-08), as described by the OECD (2017b), identifies the following groups of professionals as working in STEM professions: (1) science and engineering professionals, (2) health professionals, (3) information and communications technology professionals, and (4) science technicians and associate professionals.

On the contrary, narrower approaches focus on STEM fields, excluding health and medical professions from the STEM operational definition (Rottinghaus et al., 2018). For example, Tornese and Lupiañez-Villanueva (2017) identified professions in the science and engineering, and information and communications technology areas as STEM, and all other jobs were identified as non-STEM. In addition, Sikora and Pokropek (2012) separated health professions from STEM by creating one category for computing, engineering, and mathematics (CEM), and another one for biology, agriculture, and health professions (BAH).

Considering the broad spectrum of STEM professions, along with the rapid scientific and technological advances in today’s society (Davies & Horst, 2016; Freeman et al., 2019), STEM education has become a priority in educational agendas across many countries around the world (Fatourou et al., 2019; Gough, 2015; Joyce & Dzoga, 2011; National Research Council, 2011; Rocard et al., 2007). Notably, and in spite of the heightened policy attention and the ever-increasing demand for STEM employees, findings indicate decreasing proportions of students opting for and staying in STEM professions (Rottinghaus et al., 2018; Shoffner & Dockery, 2015; Ulriksen et al., 2015). In turn, this raises a need for policymakers, counsellors, educators, and researchers to understand the factors that contribute to students’ expectations to pursue STEM professions.

In the past two decades, both qualitative and quantitative research studies have focused on studying the predicting factors of pursuing STEM professions. However, findings cannot point to a specific combination of factors because pursuing STEM professions constitutes a complex process (Henriksen et al., 2015). For example, when examining psychological factors, research suggests that it is not only students’ cognitive abilities, such as mathematics and science performance (Hyde et al., 2008; Lindberg et al., 2010), that appear to be key factors of future academic and professional involvement in the STEM fields, but also the spatial thinking abilities and the spatial ability levels of students, among others (Gilligan-Lee et al., 2022; Hawes et al., 2017; Hegarty, 2014; Kell et al., 2013; Newcombe, 2010; Newcombe & Frick, 2010; Uttal et al., 2013).

When examining factors linked to students’ motivation for, engagement with, and learning interest towards STEM, research suggests that individual characteristics (e.g., age, attainment, and, critically, gender), family factors (e.g., home resources, parental support), school factors (e.g., teaching styles, school type, curriculum), and/or contextual and social factors (e.g., the widely held images of science and scientists) should also be considered (for an overview, see Henriksen et al., 2015). In addition, self-beliefs and self-efficacy in STEM (Lent et al., 2010), science identity and politics of recognition (Avraamidou, 2020), and the sense of belonging in STEM fields (Dortch & Patel, 2017), are associated with students’ motivation for, engagement with, and learning interest towards STEM.

Factors like the aforementioned, which are consistently being shaped by the wider social context in which students live and interact, form the basis of major theories about educational choice as well as achievement motivation, such as the expectancy-value theory of achievement motivation (EVT; Eccles, 1983; Wigfield & Eccles, 2000) and the social cognitive career theory (SCCT; Lent et al., 1994). These theories were used to guide the selection of variables for our study of student career expectations. In the EVT, the choices related to achievement in a domain are driven by individual expectations for success and subjective task value; the latter encompasses the importance of doing well on a task, which may also relate to the individual’s identity (attainment), personal interest and enjoyment (intrinsic), perceptions about the usefulness of a task (utility), and costs. The SCCT choice model postulates that choice goals for a career, as well as actions to implement those goals, are influenced by career interests, cognitive factors like perceived self-efficacy and outcome expectations (e.g. status, money), demographic and background characteristics, learning experiences, and environmental supports (Lent et al., 1994).

When aiming to study and enhance students’ interest in and motivation towards pursuing STEM professions, age needs to be taken into consideration (Mangu et al., 2015). The age of 11–15 years is considered a critical period for cultivating students’ interest in STEM careers, and for building self-efficacy with respect to mathematics and science. Notably, findings point towards the importance of intervention in even younger ages (e.g., Tai et al., 2006) arguing that students’ self-beliefs and attitudes are not equally malleable after the age of 14 (Lindahl, 2007).

Jeffries et al. (2020) examined 15-year-old students’ expectations to enrol in STEM courses at the age of 17–18 in Australia. To do so, they used data from PISA 2006 in combination with data on course choices in the last year of high school from the Longitudinal Surveys of Australian Youth. Results showed associations of demographic background, attitudes towards science, enjoyment of science, self-concept in science, and achievement in science and mathematics at the age of 15 with students’ actual choices of enrolment in STEM courses at age 17–18. Specifically, low socioeconomic status, native, and female students were less likely than their higher socioeconomic status, immigrant, and male peers to enrol in a STEM course, and these relationships were mediated by attitudes towards science and achievement in science and mathematics.

Sikora (2019) also investigated how career expectations of 16-year-olds might influence their subsequent subject choices in high school and university. To measure students’ choices, Sikora created three categories: (1) non-science choices; (2) biological, agricultural, and health-related science (BAH) choices; and (3) computing, engineering, physics, and mathematical science (CEM) choices (OECD, 2016; Sikora & Pokropek, 2012; Wegemer & Eccles, 2019). Results revealed that 35% of adolescent females expected to follow science careers, with 30% of them aiming to work in BAH but only 5% aiming to work in CEM, and that 33% of males were opting for science careers, with 21% of them aiming to work in CEM and 12% aiming for BAH. At later measurement points, the range of students expecting to enter science careers declined, and results showed a significant gender gap. Specifically, 59% of females interested in a BAH and 68% interested in a CEM career at 16 years of age changed to a non-science expectation by age 23, while 71% of males changed their expectations from BAH and 63% from CEM. Results also showed that self-concept was not a strong predictor. Findings suggest that when STEM is classified into three categories, females between the ages of 16 and 26 were more likely to work in BAH, while males were more likely to work in CEM professions. Collectively, findings highlight that the way STEM is defined matters, as findings seem to vary according to the variations of the STEM conceptualisation (Manly et al., 2018).

When Miller and Kimmel (2012) used data across the 20 years of the Longitudinal Study of American Youth to examine factors associated with the selection of and enrolment in science, technology, engineering, mathematics, and medicine (STEMM) professions, they also found significant gender differences, with males being more interested in engineering and females in medical professions. Similarly, when Sikora and Pokropek (2012) used the PISA 2006 data to examine self-concept and career expectations across 50 countries, they found that science-oriented females were more likely to prefer careers in BAH and associated fields, while males primarily favoured CEM professions. Notably, when science self-concept, science performance, and socioeconomic status were accounted for in the career plans model, in a few countries classified as transforming and developing, including Greece, females’ likelihood to consider a career in CEM increased. However, even in these cases, males were still more likely than females to consider a CEM career.

To sum up, these relatively conflicting findings underscore the importance of further research to understand the factors predicting STEM career decisions, particularly as STEM fields are classified differently across studies, leading to inconsistent conclusions regarding factors related to career preferences in these fields.

Context and Aim of the Study

Greece constitutes a noteworthy case due to the strong emphasis that has traditionally been placed on STEM education by the Greek Ministry of Education, Religious Affairs, and Sports. The Greek education system is highly centralised, with students across all schools being taught under a common curriculum and common textbooks. In lower secondary education (grades 7 to 9), physics, chemistry, biology, mathematics, geology-geography, technology, and information technology are mandatory subjects for all students (Eurydice, 2023a). Similarly, in general upper secondary education (grades 10 and 11), physics, chemistry, biology, algebra, geometry, and information technology applications are mandatory subjects for all students; at grade 11, depending on the selected specialisation, some students also have access to advanced physics and mathematics. At grade 12, and, again, depending on the selected specialisation, students have access to advanced physics, chemistry, biology, mathematics, advanced mathematics, and information technology (Eurydice, 2023b).Footnote 3

Over the last few years, the Greek Ministry of Education, Religious Affairs, and Sports has demonstrated a heightened interest in STEM education, with STEM-specific subjects and robotics becoming part of the curriculum from kindergarten to secondary school. A significant budget has been devoted to equipping schools and training teachers for the purposes of these curriculum updates (Ministry of Education, Religious Affairs, and Sports, 2023). Additionally, the ministry has put together a bank of STEM-related activities that can be used with students across different grade levels available to teachers to support them with the implementation of these innovative aspects of the curriculum (Institute of Educational Policy, 2022). Despite this increasing interest and investment in STEM, relevant targets regarding the desirable impact that such initiatives are meant to have on students’ academic and personal development and the workforce, more broadly, are still not in place.

In this context of growing interest and investment in STEM education and given the lack of guidelines as to how certain STEM initiatives can be implemented and evaluated, it would be of value to examine Greek students’ STEM career expectations and the factors that can predict them in an effort to enrich the relevant literature and inform future policies. Drawing upon existing findings (Holmegaard et al., 2014; Jeffries et al., 2020), as guided by theories of achievement motivation and career choice (EVT; Wigfield & Eccles, 2000; SCCT; Lent et al., 1994), we examined how contextual, cognitive, and non-cognitive factors, as derived from PISA 2015 (OECD, 2017c), might predict Greek 15-year-old students’ expectations for STEM careers. The research questions in the current study are:

  1. 1.

    What proportion of Greek 15-year-old students expect to follow a STEM career in their 30s?

  2. 2.

    What factors are linked with Greek 15-year-old students’ expectations to follow a STEM career in their 30s?

  3. 3.

    Do Greek 15-year-old students’ STEM career expectations and the factors linked with these expectations vary according to the way STEM is defined?

To address these research questions, we followed two approaches: (i) the wide approach, where we classified STEM professions following the ISCO-08 as utilised within PISA (OECD, 2017b), and (ii) the narrow approach, where we excluded health and medical professions from the STEM category. We also examined the extent to which the relationships between the examined factors and STEM career expectations are shaped based on the career classification used. To the best of our knowledge, there is no study to answer these research questions for the Greek population.

Methods

Participants

PISA is an international large-scale assessment conducted by the OECD to measure 15-year-old students’ knowledge and skills that allow them to participate fully in society (OECD, 2017a). It has been conducted in a 3-year cycle (which will change to a 4-year cycle from 2025), and each year it focuses on a particular domain (i.e., reading, mathematics, or science). In this study, we decided to proceed with the 2015 cycle as this is the latest cycle in which science was the major assessment domain and in which contextual questionnaires focused on science-related factors that are of interest to our study.

To select participants within each country, PISA follows a two-stage stratified sample design. In the first stage, schools are sampled with probabilities proportional to size within each participating country. In the second stage, a number of students within each sampled school are randomly selected to participate in the study (OECD, 2017c). Participation in PISA is voluntary; hence, students and their parents are given the option of withdrawing from the study. In Greece, a nationally representative sample of 5532 15-year-old students took part in the 2015 cycle and the vast majority of them attended grade 10. Of these 5532 students, 4910 had data on the main outcome of the analysis in this study and comprised the analysis sample.

Measures

Outcome Variables

Students’ career expectations constituted our main outcome variable. Students’ responses to the open-ended question “What kind of job do you expect to have when you are about 30 years old?” were coded to four-digit ISCO codes by the OECD (OECD, 2017b). Using these four-digit ISCO codes, we identified student career expectations as STEM or non-STEM based on two different approaches:

  • According to the first approach (wide), also used by the OECD (2017b), the following job families were identified as STEM: (1) science and engineering professionals; (2) health professionals; (3) information and communications technology professionals; (4) science technicians and associate professionals (for individual job titles included in each job family, see Supplementary Material at OSF: https://osf.io/9zxvh/), and all other jobs were identified as non-STEM;

  • According to the second approach (narrow), as described in the European Union (EU) STEM4YOU(th) (Tornese & Lupiañez-Villanueva, 2017), the following job families were identified as STEM: (1) science and engineering professionals; (2) information and communications technology professionals; (3) science and engineering associate professionals; (4) information and communications technicians (for individual job titles included in each job family, see Supplementary Material at OSF: https://osf.io/9zxvh/), and all other jobs were identified as non-STEM.

The main difference between these two approaches is that health professions (e.g., medical doctors, nurses) are considered STEM careers based on approach 1 (wide) but are considered non-STEM careers based on approach 2 (narrow). In our analysis, we used two binary variables (one for each approach), with STEM and non-STEM career expectations being the two categories, coded as 1 and 0, respectively.

Predictor Variables

We included student gender, family economic, social, and cultural status, immigration status, and language in the analysis to account for the student demographic background in the prediction of student career expectations. Gender as reported by students themselves is a binary variable in PISA 2015 (i.e., female/male). Family economic, social, and cultural status is a continuous index constructed by the OECD using the indicators of parent education and occupation, and home possessions, including books at home (with a Cronbach’s alpha coefficient of 0.70 for the Greek sample). Student immigration status is a categorical index comprised of the following categories: (i) native students, (ii) second-generation immigrant students, and (iii) first-generation immigrant students. Using students’ responses related to the language they usually speak at home, the OECD created an internationally comparable variable, used in this study, with the following categories: (i) language at home is the same as the language of assessment for that student (here, Greek) and (ii) language at home is another language (OECD, 2017c).

A number of indices related to students’ self-beliefs, attitudes, and motivation were also included in the analysis. Most questionnaire items related to students’ self-beliefs, attitudes, and motivation were designed by the OECD with the intention to be combined in order to measure latent constructs that could not be observed directly. For the indices of enjoyment of science, achievement motivation, instrumental motivation, test anxiety, and epistemological beliefs, students responded to 4-point Likert items to indicate the extent of agreement or disagreement with each statement. The OECD used students’ responses on five items for the enjoyment of science index (e.g., I generally have fun when I am learning broad science topics), five items for the achievement motivation index (e.g., I want to be the best, whatever I do), four items for the instrumental motivation index (e.g., Making an effort in my school science subjects is worth it because this will help me in the work I want to do later on), five items for the test anxiety index (e.g., I get very tense when I study for a test), and six items for the epistemological beliefs index (e.g., Good answers are based on evidence from many different experiments). For the self-efficacy in science index, students provided their responses to eight Likert-type statements (e.g., Describe the role of antibiotics in the treatment of disease) with four response options ranging from I could do this easily to I couldn’t do this. For the interest in broad science topics index, students provided their responses to five Likert-type statements (e.g., The Universe and its history) with four response options ranging from not interested to highly interested, and were also given a fifth option to indicate that they did not know to what the topic referred. Full lists of the items used for each of these indices along with parameter estimates can be found in OECD (2017c). All indices reached acceptable reliability for the Greek sample with all Cronbach’s alpha coefficients being above 0.73 (OECD, 2017c). All these indices have been constructed and standardised by the OECD to have an OECD mean score close to zero and an OECD standard deviation close to one. Most of the scores ranged between minus and plus two. Scores lower than the OECD average do not necessarily indicate low levels of self-beliefs or motivation or less positive attitudes; rather, they indicate self-beliefs, attitudes, and motivation that are lower than the average across the OECD countries (OECD, 2017c).

Finally, we used student achievement scores on the PISA science, mathematics, and reading literacy tests to examine the extent to which students’ career expectations are associated with their achievement. To assess students’ achievement in the three domains, PISA uses the concept of literacy. Literacy refers to “students’ capacity to apply knowledge and skills in key subjects, and to analyse, reason and communicate effectively as they identify, interpret and solve problems in a variety of situations” (OECD, 2017a, p. 13). Given that the length of the PISA test across all domains is limited to 2hours, each student is administered only a subset of test items from the total item pool. To generate population-level achievement estimates, the imputation methodology of plausible values is used. Plausible values constitute random numbers drawn from the distribution of scores that could be reasonably assigned to each student (Wu, 2005). PISA 2015 estimates ten plausible value achievement scores for each examinee. Student achievement in each PISA domain is reported on a scale that in the first cycle of PISA was set to have an international mean of 500 and a standard deviation of 100.

Data Processing and Analysis

Prior to the description of the data processing and analytical procedures we followed for the purposes of the current study, it is worth noting that thorough checking and cleaning of the PISA data are conducted by the OECD and the national teams responsible for administering PISA in each participating country before the PISA database becomes publicly available. These procedures include some initial pre-processing exploring the completeness and accuracy of the data, and they gradually move to the integration, harmonisation, and validation of the data (for additional information about each of these procedures, see OECD, 2017c).

In our results, we provide descriptive statistics related to the students’ demographic background, achievement, and other contextual characteristics, and bivariate analysis estimates for the zero-order relationships between these variables.Footnote 4 In addition, we report relevant effect sizes to demonstrate the practical significance of the examined relationships (J. Cohen, 1988). Finally, we built two hierarchical binary logistic regression models to examine the contribution of each of the selected predictor variables in the prediction of the outcome variables with other predictor variables held constant. The International Association for the Evaluation of Educational Achievement (IEA) International Database Analyzer (IDB Analyzer) (IEA, 2022) was used for all the analyses. The use of the IEA IDB Analyzer ensured that all sampling weights, replicate weights, and plausible values were used in the analysis to account for the complex nature of the PISA data and facilitate the estimation of unbiased standard errors and, subsequently, minimise the risk of type I errors (OECD, 2009; von Davier et al., 2009). The code for recoding the outcome variable for career expectations under the two approaches and a link to the PISA data for this study can be found at OSF: https://osf.io/9zxvh/.

Results

Descriptive Statistics

As can be seen in Table 1, our sample was almost equally distributed between female and male students, most students were native students, and the majority of students spoke the language of the assessment (i.e., Greek) at home. Based on the wide approach of identifying STEM and non-STEM career expectations, 28.7% of students reported that they would expect to have a STEM job at the age of 30, while based on the narrow approach, the corresponding percentage was 18.9%. This indicates that the health professionals group (e.g., medical doctors, nurses) accounts for almost 10% of the students who provided their responses to the career expectation question. In both cases, the percentages of Greek students expecting to follow STEM careers were slightly lower than the OECD averagesFootnote 5 (30.0% for approach 1 [SE = 0.07]; 20.3% for approach 2 [SE = 0.06]), with the difference being statistically significant for approach 2 (narrow), but not for approach 1 (wide).

Table 1 Numbers and percentages of students by gender, immigration status, language spoken at home, and by STEM career classification approach
Full size table

The mean in the family economic, social, and cultural status index for the analysis sample in Greece was − 0.08 (SE = 0.03), which is lower, though not statistically significantly, than the OECD average of − 0.04 (SE < 0.01). The lower mean in the index indicates that students in Greece have relatively lower family economic, social, and cultural status than their peers across the OECD countries.

In Table 2, we present students’ mean scores and related variation estimates in science, mathematics, and reading. The scores of students included in the analysis sample in this study across all three subjects were statistically significantly lower than the corresponding OECD average scores.

Table 2 Mean scores and variation in science, mathematics, and reading achievement
Full size table

In Table 3, we present students’ means and associated standard errors on each of the indices related to self-beliefs, attitudes, and motivation. Negative mean values in these indices do not necessarily indicate negative self-beliefs, attitudes, and motivation levels for the Greek sample; they indicated that students in Greece provided less positive responses compared to the OECD averages. Greek students’ levels of achievement motivation, epistemological beliefs, self-efficacy, and test anxiety were statistically significantly lower than the corresponding OECD averages. At the same time, students in Greece reported statistically significantly higher levels of enjoyment of science, instrumental motivation, and interest in broad science topics than their peers in the OECD countries, on average.

Table 3 Means and standard errors on self-beliefs, attitudes, and motivation indices
Full size table

Bivariate Statistics

In Table 4, we present the percentages of students belonging to the STEM and non-STEM career expectation categories by gender, immigration status, and language spoken at home. Based on the first approach (which includes health professions in the STEM category), the percentages of female and male students reporting that they would expect to have a STEM career at the age of 30 were similar. However, in the second approach (which does not include health professions in the STEM category), there was a statistically significant gender difference, with males being 2.5 times more likely to report that they would expect to have a STEM career at the age of 30 than females. This difference indicates that the health professionals group (e.g., medical doctors, nurses), which is considered STEM based on approach 1 and non-STEM based on approach 2, is to a great extent made up of female students.

Table 4 Percentages of students in the STEM and non-STEM career categories by STEM career classification approach and by gender, immigration status, and language spoken at home
Full size table

Students’ immigration status was not, for the most part, associated with their STEM career expectations. As can be seen in Table 4, the only statistically significant difference among the different immigration status categories was that between native and second-generation immigrant students in approach 1, with native students being more likely to report that they would expect to have a STEM-related career at the age of 30.

Similarly, the language students spoke at home more frequently was not associated with their STEM career expectations. Percentages of students belonging to the STEM career and the non-STEM career categories, respectively, were similar among students reporting that they usually speak Greek at home (i.e., the language of the assessment in Greece) and those reporting that they usually speak another language at home.

As presented in Table 5, students reporting that they would expect to have a STEM-related career at the age of 30, across both approaches, had statistically significantly higher family economic, social, and cultural status, higher levels of enjoyment of science, instrumental motivation, interest in broad science topics, self-efficacy in science, and test anxiety, and more positive epistemological beliefs. Students reporting that they would expect to have a STEM-related career at the age of 30 based on the wide approach also had statistically significantly higher levels of achievement motivation but this difference was not statistically significant in the narrow approach. Across both approaches, the largest effect sizes were noted for the enjoyment of science and instrumental motivation indices, while all the differences yielded larger effect sizes in the wide compared to the narrow approach, indicating larger differences in scores among students belonging to the STEM and non-STEM career categories when health professionals (e.g., medical doctors, nurses) are considered STEM-related careers (approach 1) compared to when they are considered non-STEM careers (approach 2).

Table 5 Means of students in the STEM and non-STEM career categories on the family economic, social, and cultural status and the self-beliefs, attitudes, and motivation indices, and in science, mathematics, and reading by STEM career classification approach
Full size table

Across both approaches, students reporting that they would expect to have a STEM-related career at the age of 30 had statistically significantly higher science, mathematics, and reading scores compared to their peers who did not expect to have a STEM-related career. These differences were similar in science and mathematics, and slightly smaller in reading within each of the approaches, while all differences yielded much larger effect sizes in the wide compared to the narrow approach, indicating larger differences in scores among students belonging to the STEM and non-STEM career categories when health professionals (e.g., medical doctors, nurses) are considered STEM-related careers (approach 1) compared to when they are considered non-STEM careers (approach 2).

Binary Logistic Regression Models

To examine the contribution of each one of the selected variables in the prediction of Greek students’ career expectations, while accounting for other factors, we constructed two hierarchical binary logistic regression models. In Tables 6 and 7, we summarise the results of these models for the wide and the narrow STEM career classification approach, respectively.

Table 6 Hierarchical binary logistic regression model for STEM career expectations, approach 1 (wide)
Full size table
Table 7 Hierarchical binary logistic regression model for STEM career expectations, approach 2 (narrow)
Full size table

Approach 1 (Wide)

In the final step of the approach 1 (wide) model, STEM career expectations were statistically significantly associated with students’ family economic, social, and cultural status, their enjoyment of science, instrumental motivation, interest in broad science topics, and their science performance (Table 6). The most robust predictor of STEM career expectations, even after accounting for a range of demographic, cognitive, and non-cognitive factors, was students’ instrumental motivation. With every extra unit in the instrumental motivation index, students’ odds of belonging to the STEM career compared to the non-STEM career category were increased by 1.93 times. This indicates that the more students perceived studying science in school as useful to their future lives and careers, the more likely they were to report that they would expect to have a STEM-related career at the age of 30 compared to their peers who did not perceive studying science in school as useful to their future lives and careers or did so to a lesser extent.

The next most robust predictors of STEM career expectations were students’ enjoyment of science, their interest in broad science topics, and their family’s economic, social, and cultural status. With every extra unit in the enjoyment of science index, students’ odds of belonging to the STEM career compared to the non-STEM career category were increased by 1.23 times, and with every extra unit in the interest in broad science topics and family economic, social, and cultural status indices, students’ odds of belonging to the STEM career compared to the non-STEM career category were increased by 1.17 times, respectively.

Students’ science score on the PISA test was also a statistically significant predictor of their STEM career expectations. An increase of 100 points in students’ science score (which is approximately one standard deviation) was expected to increase their odds of belonging to the STEM career as opposed to the non-STEM career category by 49%.Footnote 6 Although achievement motivation, epistemological beliefs, self-efficacy in science, and test anxiety were originally statistically significantly related to students’ career expectations at a bivariate level, they were no longer statistically significant predictors of students’ STEM career expectations when they were examined within a regression context.

The Nagelkerke index of R2 indicated that the final model explained 21% of the variance in the outcome variable. Although this proportion of explained variance is considerable, it is acknowledged that additional and/or different variables might be relevant in explaining the remaining unexplained variance.

Approach 2 (Narrow)

In the final step of the approach 2 (narrow) model, STEM career expectations were statistically significantly associated with students’ gender, achievement motivation, enjoyment of science, instrumental motivation, interest in broad science topics, and their science performance (Table 7). Family economic, social, and cultural status, although a statistically significant predictor of students’ STEM career expectations in steps 1 and 2 of the model, lost its statistical significance when students’ science performance was entered into the model.

The most robust predictor of STEM career expectations, even after accounting for a range of demographic, cognitive, and non-cognitive factors, was students’ gender. Male students’ odds of reporting that they would expect to have a STEM-related career at the age of 30 were 3.09 times higher compared to those of female students. This gender difference retained its statistical significance across all steps of the model and did not lose any of its robustness when other variables were entered into the model.

The next most robust predictor of STEM career expectations was students’ instrumental motivation. With every extra unit in the instrumental motivation index, students’ odds of belonging to the STEM career compared to the non-STEM career category were increased by 1.39 times. This indicates that the more students perceived studying science in school as useful to their future lives and careers, the more likely they were to report that they would expect to have a STEM-related career at the age of 30 compared to their peers who did not perceive studying science in school as useful to their future lives and careers.

Students who reported being more motivated to achieve —in both school and beyond, tended to be less likely to report that they would expect to have a STEM-related career at the age of 30. Specifically, with every extra unit in the achievement motivation index, the change in the odds of students belonging to the STEM career compared to the non-STEM career category was 0.77.Footnote 7 Students who reported higher levels of enjoyment of science and interest in broad science topics were more likely to report that they would expect to have a STEM-related career at the age of 30 compared to their peers. With every extra unit in the enjoyment of science and interest in broad science topics indices, students’ odds of belonging to the STEM career compared to the non-STEM career category were increased by 1.13 and 1.21 times, respectively.

Students’ science score on the PISA test was also a statistically significant predictor of their STEM career expectations. An increase of 100 points in students’ science score (which is approximately one standard deviation) was expected to increase their odds of belonging to the STEM career as opposed to the non-STEM career category by 22%.Footnote 8 Although family economic, social, and cultural status, epistemological beliefs, self-efficacy in science, and test anxiety were originally statistically significantly related to students’ career expectations at a bivariate level, they were no longer statistically significant predictors of students’ STEM career expectations when they were examined within a regression context.

The Nagelkerke index of R2 indicated that the final model explained 14% of the variance in the outcome variable. Although this proportion of explained variance is not negligible, it is acknowledged that additional and/or different variables might be relevant in explaining the remaining unexplained variance.

Discussion

Involvement with STEM subjects during schooling has been researched and increasingly promoted across different education systems (Blasko et al., 2018). Modern curricula prioritise skills like problem-solving and critical thinking, and STEM education programmes can foster such complex skills. Given the promise for advancing innovation and entrepreneurship in a technology-oriented environment, and the anticipated job growth in science, engineering, and information and communications technology fields, it is unsurprising that students and their parents, as well as policymakers and governments, value STEM careers and pursue training in these domains. In this paper, we examined the self-reported expectations for STEM and non-STEM careers of a representative sample of Greek 15-year-olds considering the contribution of a range of demographic, cognitive, and non-cognitive variables in the prediction of these expectations, along with some implications for educational policy and practice.

Based on the wide approach of identifying STEM and non-STEM career expectations, 28.7% of students (with about a third of that attributed to the health and medical professions) in our sample reported that they would expect to have a STEM job at the age of 30, while based on the narrow approach, the corresponding percentage was 18.9%. Most of the examined variables were found to be significantly related to STEM career expectations in at least one of the two STEM career classification approaches within a bivariate context. However, when all demographic, cognitive, and non-cognitive variables were examined simultaneously in predicting STEM career expectations at age 30, more concise patterns were detected.

In the first approach, with health and medical professions included in the STEM category, instrumental motivation for learning science was the strongest predictor of STEM career expectations, after accounting for other factors in the model. This indicates that students’ agreement with statements about the importance and usefulness of studying science for future career prospects is a powerful predictor of their STEM career expectations, which matches the utility value in EVT (Wigfield & Eccles, 2000). Other motivation variables (i.e., enjoyment of science and interest in broad science topics) that resemble EVT’s intrinsic value (Wigfield & Eccles, 2000) and interest and values in SCCT (Lent et al., 1994) significantly predicted STEM career expectations, but to a lesser degree. However, achievement motivation and self-efficacy in science were not found to predict STEM career expectations. Dimensions of student attainment value and, especially, expectancies for success are important in EVT for predicting achievement or task success (Lee & Stankov, 2018; Michaelides et al., 2019). This does not seem to be the case when examining STEM career expectations presumably because the mechanisms are different compared to achievement models. Also, along these lines, neither epistemological beliefs nor test anxiety were significant predictors of STEM career expectations within the wide approach model.

In accordance with existing literature that has shown that socioeconomic status plays a role in career choices (Sikora & Pokropek, 2012), the current findings revealed family’s economic, social, and cultural status as the single demographic characteristic that predicted STEM career expectations, with other variables held constant. PISA science achievement was also a significant predictor of STEM career expectations. SCCT’s choice model (Lent et al., 1994) considers background influences and affordances as shaping students’ learning experiences, and the overall environment as supportive or unsupportive for choice-making. Within this context, though, it is important to note that the overall model had a modest explanatory power, and most of it was attributed to the motivational variables in the wider approach.

Focusing on the narrower definition of STEM professions that excludes the health and medical professions from the STEM category, we fitted the same hierarchical binary logistic regression model for predicting STEM career expectations. The explanatory power of this model was lower compared to the wide approach model. Consistent with the first model, instrumental motivation, enjoyment of science, and interest in broad science topics as well as PISA science achievement were positively associated with STEM career expectations, highlighting the importance of these motivational factors in career choices regardless of the career classification used.

Despite these similarities between the two models, there were also some noteworthy differences. In particular, although family’s economic, social, and cultural status had a significant relationship with students’ career expectations within a bivariate context across both STEM career classification approaches, supporting SCCT’s argument that social class can be a contextual influence for choice (Lent et al., 1994), suggesting a role for intergenerational influences in career choices, this was not the case when all predictors were included in the same model in the narrow approach. A potential explanation can be supported by the descriptive statistics in Table 5; the difference between the STEM and non-STEM average in the family’s economic, social, and cultural status variable was twice as large in approach 1 (wide) compared to approach 2 (narrow). This large discrepancy between the approaches could be attributed to the students reporting health and medical professions, who are not classified as STEM in the second approach. It appears that, on average, these students who expect to work in STEM careers in the wide but not the narrow approach come from higher socioeconomic backgrounds. Moreover, jobs under the “ICT professionals” classification appear in both approaches, but “ICT technicians” are included only in the narrow approach (see Supplementary Material). These two differences in the STEM career classification could potentially explain why the family’s economic, social, and cultural status is significant in the first but not in the second approach.

Another difference between the two models was the association of students’ motivation to achieve well in school and beyond with STEM career expectations; while in the wide approach model, the variable was not associated with STEM career expectations, an unexpected negative association was found in the narrow approach model after controlling for covariates, and this negative relationship was found to be stronger for boys than girls. The SCCT’s choice model might be helpful in interpreting this finding. Choice goals and actions in SCCT seemed to be shaped by interests, outcome expectations, self-efficacy beliefs, learning experience, and demographic background (Lent et al., 1994), but not by achievement motivation.

A final difference between the two models was how gender emerged as a significant predictor in the narrower approach, with males being more likely than females to report STEM career expectations. Males and females show differential preferences for CEM versus BAH choices (Sikora & Pokropek, 2012). This finding confirms previous research evidence (Manly et al., 2018) and highlights the importance of STEM career classification approaches in findings related to career expectation predictors. In professions related to mathematics, physics, technology, and engineering, a noticeable gender gap exists. In line with existing studies (e.g., Miller & Kimmel, 2012; Sikora, 2019), this gap does not exist in the wider STEM definition, i.e., when ICT technicians are excluded from the STEM category and when medical and health professions are included because there is a stronger preference from females for the latter. The different results obtained under the two approaches suggest that relevant research needs to be transparent about definitions and operationalisations of STEM. We would argue that the STEM versus non-STEM dichotomy, along with the associated stereotypes, can often be too broad and finer distinctions can be identified and appropriately reported. Generalisations about the role of certain factors in career expectations and choices, and recommendations about proposed policies should be accompanied by exact definitions of STEM.

One limitation of the present study is the relatively modest explanatory power of the models despite the large number of demographic, cognitive, non-cognitive predictors considered. Given that we were limited to the information available in the PISA database, it would be advantageous for future relevant studies to examine additional predictors, such as parental or teacher beliefs regarding a student’s educational and vocational choices, educational opportunities and experiences, encompassing early exposure to diverse or contemporary subject areas and occupational possibilities, and factors related to students’ plans for their subsequent careers (e.g., out-of-school lessons). Even though results based on data from international large-scale assessments, including PISA, that draw on nationally representative samples, have the potential to be scalable and transferable, their non-experimental nature does not allow for the establishment of causal relationships among the examined variables (L. Cohen et al., 2017). Also, inferences about the relationships between the predictor variables, especially the ones related to students’ self-beliefs, attitudes, and motivation, and the outcome variables should consider that the relationships may be reciprocal.

Interestingly, despite their limitations, our findings are in line with experimental findings suggesting that the formation of students’ future career aspirations is malleable and influenced by environmental factors. For example, Barone et al. (2019) demonstrated that they could alter career decision-making by increasing counselling activities focusing on occupational prospects. Familiarity with career options can interact with students’ instrumental motivation, promote the development of clearer career goals, and facilitate planning by enabling students to select relevant coursework and engage in suitable activities early in their academic trajectory. Lastly, it should be noted that the primary outcome variable of this study was 15-year-old students’ career expectations at age 30. Future research endeavours should extend to longitudinal designs and incorporate additional outcome variables, such as university degree choices and employment after graduation, to capture individuals’ actual participation in STEM, and other, fields.

Altogether, our findings have significant implications for STEM research and educational policies on STEM within the Greek context. A key finding that could be used by policymakers is the need for transparency in definitions and operationalisations of STEM, as this seems to be associated with the purported role of important factors, such as gender, for students’ career expectations in these fields. Despite the mandatory STEM-related courses that have long existed and more recent steps towards a more STEM-focused curriculum (Eurydice, 2023b), our findings suggest that there is still room for improvement within the Greek education system, e.g., STEM-related counselling initiatives to familiarise students with STEM-related careers (Barone et al., 2019). Based on our findings, enhancing Greek students’ instrumental motivation and science achievement seems to be crucial for increasing the uptake of STEM-related courses and STEM careers. Attempts to enrich specific content areas like spatial thinking in the school curriculum, while focusing on activities aimed at enhancing spatial skills (linked to STEM motivation and achievement according to Newcombe, 2017), hold the potential to boost students’ instrumental motivation and science achievement and, in turn, inspire them to pursue STEM careers (Newcombe, 2013; Zhu et al., 2023). Lastly, policymakers should be aware that there are socioeconomic differences within the broader STEM career expectations; as we have shown, there is variation in the categories of medical and ICT technician jobs. This emphasises the need for the Greek education system to provide guidance and funding for inclusive STEM initiatives, combined with any existing efforts to improve STEM equipment and resources in Greek schools, to address the needs of all students, including those coming from lower socioeconomic backgrounds (Adams et al., 2018).