3.1 Method

To answer the research question and using the literature review as support, we developed a model enabling us to predict the impact of explanatory variables on the probability of attaining an education-job match for Portuguese higher education graduates. The dependent variable was the graduate education-job match. This variable was dichotomous and indicated whether a match existed between the field of study and the type of work being undertaken. Given the qualitative nature (0 or 1) of the dependent variable, the use of a logistic regression model was recommended. This was in accordance with the method used by Frenkel and Leck (2017) and Garcia-Espejo and Ibanez (2006).

The independent variables related to the hypothesis of this study, being decision variables, include the institutional characteristics of the Portuguese binary higher education system, or university and polytechnic education.

The group nominated as control variables included sociodemographic determinants such as gender, age and mobility for work (commuting to a different district to work), as well as academic trajectory factors, such as the final grade, the year of completing the degree, mobility for studies (travelling to a different district to study), the field of study (education; arts and humanities; social sciences, commerce, and law; natural sciences, mathematics and statistics; engineering, manufacturing and construction; agriculture; health and welfare; and services) and participation in a set of extracurricular activities (international mobility, participation in associations, volunteering, complementary training, internships).

Thus, we estimated the function as follows:

where Y was a binary dependent variable that took the value 1 or 0, X was the matrix of explanatory variables and $\epsilon$ was the unobserved component or statistical error.

In probabilistic terms, we used the following equation:

Where $\beta$ were the parameters to be estimated and X were the explanatory variables. Assuming that F (X) undertook the logistic cumulative distribution function, we calculated it as follows:

where i = 1 if there was a probability of the education-job match being attained and i = 0 otherwise.

3.2 Data and variables

The sample consists of the graduates of a university and a polytechnic institute, both located in the northern region of Portugal, who obtained their degrees between 2014 and 2019. The location was chosen because it is the most populated region of Portugal and it has the highest employed population, comprising 35% of the employed population of Portugal. This is also the Portuguese region with the second highest number of graduates, comprising 33% of the country’s total. The HEIs considered were randomly chosen, and only those who returned an acceptable number of responses were used to further develop the research.

The data was collected from June 2020 to October 2020 with an online questionnaire, which was pretested twice for good adjustment to the model. This was done firstly by face-to-face interviews with a group of volunteer graduates to check the adequacy of the instrument, followed by an experimental collection of information from the results of graduate surveys from an university not included in the sample.

The questionnaires were divided into three major groups of questions, one of which asked about personal information, including personal and family characteristics of graduates. Another group covered academic training, with a set of questions about the academic trajectory of graduates. The final group enquired about how graduates integrated into the labour market, as well as their opinions.

The HEIs sent questionnaires by email to their own graduates due to data protection regulations. The questionnaire was also posted on social networks with an access link. This method of disseminating the questionnaires prevents us from finding out exactly how many questionnaires were sent out.

We obtained 694 eligible answers, of which 246 corresponded to graduates who were not working and 448 to those who were, the latter group of which was the object of analysis in this study. To analyse the data collected, according to the research question and hypothesis formulated, and in addition to the statistical descriptive analysis, a logistic regression model was estimated, using the variables described below.

The dependent variable was dichotomous and indicated whether there was a match between the field of study and the type of employment obtained. We could have had a different dependent variable, such as employability, motivated by the growing interest of researchers in understanding how graduates’ transition to the labour market. However, this variable would have only considered whether graduates were employed or not. We wanted to go further and study the education-job match of graduates, that is, the adjustment between the skills required to work and the knowledge acquired from their qualifications. Supported in the literature (e.g., Frenkel & Leck, 2017; Garcia-Espejo & Ibanez, 2006), we believe that this is the most appropriate approach for a match between graduates’ training and their job roles to be identified.

According to Hartog (2000), there are three approaches for measuring an education–job match: (1) job analysis, which is based on an ‘objective’ evaluation by professional job analysts; this approach is conceptually better, but its measurement is generally not easy to obtain; (2) realised matches, based on the observed educational achievement of workers in each occupation; this method may result in biases for the overeducated and the undereducated; and (3) workers’ self-assessments, where they subjectively assess their own education-job matches. The author noted that the third approach is considered the most traditional way to measure overeducation.

Our study is based on a subjective assessment that used the workers’ self-assessment approach to graduates, who responded to the following higher education survey based around the question: ‘How related is your job to your field of study?’ The graduates were asked to rate their answers on a five-point Likert scale, running from 1 to 5 points, 1 meaning “not related” and 5 being “completely related”. This was then converted into a binary variable that was used as a dependent variable in a logistic regression model, grouping scores of 1 and 2 as mismatches (0) and 4 and 5 as matches (1). A score of 3 was the neutral point of the scale, which we considered to represent a match (1). Other studies have also used this subjective assessment of workers (e.g., Boudarbat & Chernoff, 2012; Frenkel & Leck, 2017; Green & McIntosh, 2007; Iammarino & Marinelli, 2015).

We broke down the potential determinants of a match into two general categories according to the independent variables (Table 3), decision variables and control variables.

Table 4 presents the descriptive statistics for key variables, divided into graduates whose job matches the field of study and those whose job does not.

The evidence from the descriptive data shows that a high percentage of graduates (62.7%) described their job as closely related to their field of study, which means that most of them were applying the knowledge they acquired in higher education to their professions.

The data also shows that more women than men had jobs that did not match their fields of study. Age seemed to have a similar effect in both cases, in terms of informants whose jobs matched their qualifications and those who did not have a match, most of whom belonged to the 20-25 and 26-30 age ranges, predominant for both groups. A similar situation occurred for the final grade and the year the degree was completed, where both types of graduates had the same characteristics for these indicators, most of them being rewarded 14 or 15 points and finishing it in 2016 or 2017.

The fields of study related to “social science, business studies and law”, “engineering, manufacturing and construction” and “natural science, mathematics and statistics” comprised the highest percentage of graduates attaining an education-job match.

Only the frequency of complementary training and the participation in internship programs seemed to favour education-job matching, since a large proportion of the respondents stated that they had participated in them, mainly graduates who affirmed that their jobs matched their respective fields of study.

The data illustrates that more than half (about 60%) of those who needed both mobility for work (i.e., working in a district that was different from their home address) and mobility for study (i.e., having studied in a district that was different from their home address) had a successful education-job match.

Finally, there were differences regarding the type of HEI, with 54% of university graduates and 46% of polytechnic graduates considering that they had a successful education-job match. We have attempted to identify whether there is significant evidence for these differences, via a more robust analysis.1

3.3 Results and discussion

To assess whether there were structural differences in the logistic regression between the university and polytechnic subsystems, we started by estimating an unrestricted model that, in addition to the explanatory variables, included a dummy variable (D = 1 if graduate n studied at a polytechnic and D = 0 if graduate n studied at a university) with additive and multiplicative methods, doubling the number of parameters estimated. A likelihood ratio test (LR) was then applied. Chi-square = 53.92 (with 18 degrees of freedom) allowed us to conclude with a 1% significance level that there were structural differences between the two subsystems.

Table 5 includes the results of the global sample for each higher education subsystem. Based on McFadden’s R-squared, the likelihood ratio test and the number of 'correctly predicted' cases, the results show that the estimates seemed to fit well with the data collected. As can be seen, there was a significant LR, so we rejected H0, i.e., the restricted model (Model 1), leading to a need to estimate and analyse separate models for university (Model 1.A) and polytechnic (Model 1.B) graduates.

The results from Models 1.A and 1.B appear to confirm the hypothesis of this study, namely that due to the different objectives of each of the subsystems in Portuguese higher education, there were differences in the determinants, which affected the probability of attaining an education-job match for university and polytechnic graduates. Therefore, although the boundaries between universities and polytechnics are less obvious today (Queiró, 2017), both often offering similar programmes (Ferreira et al., 2008), the results may imply that the differences between the two higher education subsystems are still noted in the education-job matches of their graduates.

Among the control variables, gender determined the education-job match only in the university subsystem. This variable was negative and significantly related to the dependent variable for these graduates, which means that women were less likely to be education-job matched than men. This was corroborated in the literature, since Berntson (2008), McQuaid and Lindsay (2005) and Wilton (2008) found that men have access to better options in the labour market and are, therefore, seen as more “employable”. There is no consensus about this, however, as other studies have shown that there are no significant differences between recent male and female graduates (García-Espejo & Ibáñez, 2006; Green & McIntosh, 2007; Mavromaras et al., 2009; Menon, et al., 2012) and studies with recent Portuguese graduates have also shown that gender is not a determinant of their employability (Galego & Caleiro, 2011; Sin et al., 2016). These latest findings are similar to the results obtained for polytechnic graduates, where gender is not a significant variable.

The variable age was not significant enough to explain the education-job match of graduates from either of the higher education subsystems, contradicting the findings of Galego and Caleiro (2011) for a Portuguese university, which showed that younger graduates seemed to be in a disadvantageous position compared to more mature graduates upon entering the labour market.

Mobility for work was significant in the education-job match of graduates for both subsystems, but in opposite directions. On the one hand, the effect was negative for university graduates who had to travel to a different district for work, with a lower probability of an education-job match. The observations of Pereira (2007) can shed some light on this result to some extent, since he studied geographic mobility in Portugal using the distance of displacement as the reference domain, concluding that this mechanism was not an effective way of adjusting the labour market. On the other hand, for polytechnic graduates, commuting to a different district for work positively affected their probability of an education-job match. This may be due to the polytechnic considered in our study not being located in a metropolitan area and, as explained by Frenkel and Leck (2017), graduates from peripheral areas often have to migrate to core regions for a better chance of finding jobs matching their qualifications and skills.

The results for final grades were consistent with other studies which have shown that good grades increase the likelihood of an education-job match for graduates (Finch et al., 2013; Galego & Caleiro, 2011; Garcia-Espejo & Ibanez, 2006; Grayson, 2004; Verhaest & Omey, 2010). The relationship between this variable and the probability of this kind of match was positive and significant for both university and polytechnic graduates. The literature also shows that these results essentially make sense for recent graduates transitioning to the labour market because as individuals acquire professional experience, other sources of information are considered by employers and the final grade loses its effect in determining the probability of obtaining a job (Menon et al., 2012).

The year a degree was completed was not significant in the education-job match of graduates in either of the higher education subsystems. We introduced a trend with this variable in order to identify whether the probability of attaining such a match could change over time. This parameter was not significant, implying that it did not happen.

Mobility for study only affected the probability of an education-job match for university graduates, where it had a positive and significant impact. This result may be associated with the fact that employers believe that mobility allows more skilled individuals to cope with different organisational contexts and that it improves their performance at work (Sin et al., 2017). In general, when the literature mentions the benefits of mobility, it refers to international mobility programmes during academic studies. However, in order to explain the results of our study, we have established a parallel with this type of mobility and suggested that travelling to a different district to study develops the personality characteristics of graduates such as flexibility and adaptability in new environments and lower risk aversion, as well as interpersonal skills like autonomy, the ability to solve problems, organisation and coordination, in the same way that international mobility does (Konevas & Duoba, 2007), all of which are valued by employers in the labour market. For polytechnic graduates, this is not a significant parameter.

The fields of "social science, commerce and law", “engineering, manufacturing and construction” and “health and well-being” affected education-job matching in the university subsystem, as several other studies have found as well (Buonanno & Pozzoli, 2009; Frenkel & Leck, 2017; Galego & Caleiro, 2011). The results show that other fields also positively affected the dependent variable, such as “health and well-being” and "natural science, mathematics and statistics". The result for the latter field was attested to by the OECD data, which demonstrates that one of the main fields where Portuguese graduates were employed was in "natural science, mathematics and statistics". In the polytechnic subsystem, only graduates in the fields of “social science, commerce and law” and “engineering, manufacturing and construction” were more likely to have an education-job match, with these variables showing a positive effect on the dependent variable. Graduates in the fields of "education", "arts and humanities" and "agriculture" had a lower likelihood of an education-job match, since these variables had a negative sign and were statistically significant. We do not believe that the main reason for this stemmed from the inability of polytechnic graduates from these fields of study to integrate successfully into the labour market and find matching jobs, but rather that it was due to the differences between academic institutions, where polytechnic institutes, according to the Education Law (Law No. 46/86), have a more practical and professional nature while “arts and humanities” is more closely related to university studies, meaning that it is more academic and scientific in nature than what is offered at polytechnics.

Participation in extracurricular activities, which is considered in general by employers as a way to develop individual skills (Roulin & Bangerter, 2013), was expected to have a positive effect on education-job matching. However, the results show that it was only partially true, for both the university and the polytechnic subsystems. Out of the five extracurricular activities considered in the model, only the frequency of complementary training had a significant positive effect on the dependent variable. This result is backed up by the literature in the sense that it confirms the importance of developing soft skills from complementary training, in addition to the hard skills covered in curricular plans (Suleman, 2016; Suleman & Laranjeiro, 2018). Other extracurricular activities, namely, international mobility, participation in associations and participation in internships, did not have a significant effect on education-job matching. Volunteering was negatively related to the education-job match in the polytechnic subsystem; that is, graduates who participated in volunteer programs were less likely to find a job connected to their field of study. This may have been because some employers believed that strong involvement in extracurricular activities could compromise academic performance (Rubin et al., 2002) and lead to less professional commitment (Roulin & Bangerter, 2013), thus potentially harming education-job matching.