Why are lower socioeconomic background students underrepresented in Erasmus? A focus on the selection into mobility and degree course organization

ABSTRACT

Disadvantaged students in higher education are significantly underrepresented in study abroad programs. Existing research has explored this socioeconomic disparity in student mobility, primarily by comparing mobile and non-mobile students. However, these studies often overlook the extensive selection process students undergo to join a mobility program, encompassing application phases, eligibility assessments, and ability evaluations. This study investigates how these selection steps contribute to the unequal uptake of higher education mobility in Europe, focusing on the European Union’s most popular study-abroad program ‘Erasmus’. Additionally, it explores how the program organization at the degree course level is associated with socioeconomic disparities in student mobility. Using administrative population data from a large Italian university, covering 46,000 graduates, the study finds that, on average, 90% of socioeconomic disparities in Erasmus mobility are due to disadvantaged students shying away from the application, while ability selection is only marginally important. Additionally, degree-level characteristics, particularly the social segregation across fields of study, matter. Although the study is based on data from a single university in Italy, it has implications for an international audience in countries with similar institutional settings for international student mobility, as it demonstrates for the first time the impact of the selection process and degree course organization on the socioeconomic gap in study abroad program participation.

KEYWORDS:

• International student mobility
• Erasmus
• inequality
• selection
• unequal uptake
• Europe

1. Introduction

A rising share of students opt for international student mobility (ISM) and complete a part or all of their studies abroad. Across OECD countries, Europe is the second largest source of international students, making up 22% of all mobile students (OECD 2023). Student mobility in Europe mainly consists of credit mobility, where students spend a part of their study time abroad while remaining enrolled in their home country’s higher education institute, thanks to the popular Erasmus program.¹

From a social policy perspective, fostering student mobility abroad is a valid instrument for increasing young people’s tolerance, multiculturalism and integration into different cultural settings (Roy et al. 2019). Research shows that participating in study abroad programs is not only generally regarded as a great experience by participants but can also benefit students’ professional careers, for example their language competencies (Sorrenti 2017), their future educational pathways and employment chances (d’Hombres and Schnepf 2021; Jacob, Kuehhirt, and Rodrigues 2019) and higher wage growth after graduation (Kratz and Netz 2018).

Nevertheless, these beneficial outcomes of ISM are not equally distributed among tertiary education students, since it is especially those students with a higher parental socioeconomic background who participate in ISM (Hauschildt et al. 2015; Di Pietro 2020). The problem is universal in Europe even though there are country variations in its extent (Hauschildt et al. 2015). Consequently, some research suggests that ISM fortifies existing social stratification in European societies (Bargel et al. 2009; Findlay 2011; Netz and Finger 2016).

Policymakers try to counteract unequal ISM uptake stressing the need for inclusion (e.g. European Commission 2017). A major issue is, however, to understand which measures help increase the number of disadvantaged ISM participants. Extant literature is limited in advising on policy design since it takes a student-centered approach (Cairns 2019), stressing the importance of differences between privileged and less privileged students in terms of, e.g. language knowledge, access to information and financial resources. However, recent research shows that students’ environment, i.e. the university they attend as well as their chosen field of study, explains some part of the socioeconomic gap in ISM (Schnepf and Colagrossi 2020; Schnepf, Bastianelli, and Blasko 2022). These results indicate clearly that it is not only individuals’ choices that drive low ISM uptake of disadvantaged students.

As a consequence, the question arises whether the so-called ‘meso’-level, the institutional management of international student mobility, impacts the socioeconomic uptake gap (Cairns 2019). The value added of this study is to examine for the first time two meso-level characteristics of ISM: the selection process leading to participation in study-abroad programs (from application to ability selection to participation) and the management of ISM mobility at the finer institutional level, the single degree-course.

To do so, this paper focuses on the Erasmus mobility program, which is the most well-known European international credit mobility program. In 2019, it funded as many as 335,600 higher education students studying abroad (European Commission 2019), representing about 50% of all higher education credit mobility students in Europe (European Commission 2021, Figure 63). The study exploits a rich dataset combining administrative data on Erasmus applications with survey data on student family and school backgrounds in the second largest public university in Italy, the University of Bologna.

2. Literature, selection into Erasmus mobility and research questions

2.1. What are the mechanisms driving unequal uptake of mobility programs?

The predominant part of the literature takes a student-centered approach (Cairns 2019) focusing on individual characteristics of disadvantaged students who are generally defined to have a lower socioeconomic background in terms of parental education, occupation or income. Disadvantaged students tend to perceive ISM as less beneficial (Loerz, Netz, and Quast 2015), have lower previous mobility experiences (Brooks and Waters 2010), suffer from a lack of financial resources (Findlay 2011) and language skills (Souto-Otero et al. 2013) and are less informed (Teichler 2017) compared to advantaged students. Since all these characteristics are important for studying abroad, disadvantaged students enroll less in ISM than their better-off counterparts. Nevertheless, these characteristics can generally only explain a small part of the socioeconomic gap in mobility uptake.

Previous research has shown that the meso-level context matters as well and with this the opportunity structure for study abroad programs in students’ higher education environments. This includes the universities’ level of engagement in international mobility, their ability to negotiate ISM agreements and the perceived relevance of studying abroad (Netz 2015), as well as the financial governance of the program (Cairns 2019).² Recent research shows that the university attended as well as students’ chosen field of study also explain some part of the socioeconomic gap in ISM. For example, inequality in uptake tends to be lower for fields of study that have very high levels of ISM mobility, like language subjects (Schnepf and Colagrossi 2020). Research focusing on universities shows that disadvantaged students are more likely to be enrolled in those universities with lower ISM opportunities (Schnepf, Bastianelli, and Blasko 2022). Consequently, it is not only individuals’ choices but also their study environment, which drives low ISM uptake of disadvantaged students.

However, still very little is known about how students’ learning environment matters since meso-level factors related to the institutional management of ISM are hardly ever observed. This study aims at partly filling this gap by examining for the first time, to the knowledge of the authors, two meso-level components, the student grant allocation process and the importance of students’ degree course organization linked to mobility, to add insights to existing knowledge on social disadvantage in study-abroad program participation. Our contribution is twofold: first, we are able to measure aspects of the institutional opportunity structure in ISM that previous research recognized as relevant but was not able to measure because of data limitations. Second, we are able to focus on the institutional organizational level that is closest to students, i.e. their degree course, in contrast to existing literature that – if examining meso-level factors – can only focus on the broader level of the university or field of study.

2.2. The student grant allocation process

Existing studies, quantifying and examining the unequal student mobility uptake with country survey or population data, examine inequality in participation. However, whether a student takes part in credit mobility is the result of a number of meso-level selection steps taking place at university.

In Europe, the management of the selection process at the university level is decided by the university and, to the knowledge of the authors, there is no documentation on the specific selection criteria of credit mobile students by European universities. Nevertheless, given that universities receive only a limited amount of Erasmus mobility funding from the National Agency responsible for Erasmus (Agenzia Nazionale Erasmus + (INDIRE) for higher education in Italy) and language skills are key for studying abroad, selection criteria likely comprise ability and language skill selection throughout European universities (Diaz 2017). As such, the following selection steps in place at the University of Bologna are likely to be similar at least to other European universities.

First, a student needs to apply to Erasmus generally with a motivation letter, a study plan and information on their language skills, and hence know the ISM program and have access to information on how to apply (Diaz 2017). Once applied, second, the University needs to select those applicants who are eligible. At the University of Bologna, only sufficient knowledge of the language of teaching in the foreign university the student wants to study in (European level A2) is required. Third, universities tend to opt for ability selection among the pool of eligible candidates. At the University of Bologna, for Erasmus mobility schemes offered within the students’ fields of study, applicants are ranked based on their previous academic performance and the quality of their application. For each mobility scheme, the available mobility grants are allocated to the highest-ranked applicants; the application score of the last student in the ranking who is offered the mobility grant is the ‘threshold’ score below which students are rejected. Fourth and finally, a student who was awarded the mobility grant needs to decide whether indeed to take part in the program. Final participants obtain the EU Erasmus grant and top-ups by the Italian Ministry of Education which increase with decreasing students' household income.

These selection steps are key mechanisms of unequal ISM uptake. For example, if ability selection at universities filters most dominantly disadvantaged students from the pool of applicants, we can clearly say that ability selection needs to be revised to make Erasmus more inclusive. If, however, disadvantaged students are less informed about ISM or more deterred by the application process and its requirements (as argued in Diaz 2017), the socioeconomic pattern of the applications would already predict those of the later ISM participation.

Research question 1: At which stage of the selection process into Erasmus higher education mobility do disadvantaged students drop out: at the application phase, the eligibility screening, the ability selection or the final uptake decision?

Another important meso-level feature of Erasmus regards the organization of inter-institutional agreements with host universities. Inter-institutional agreements are generally formed across university faculties or departments and are therefore decentralized. Faculties or departments tend to organize a number of degree courses, within which students are likely to have close networks given that they follow study programs within the same broad disciplinary field. These networks could be important for ISM information exchange between peers returning and planning ISM. In addition, mobility schemes are often linked to specific degree courses and not the faculty or department as a whole. Nevertheless, a small share of mobility is open to students from any field (so called ‘interdisciplinary programs’).

Consequently, it is sensible to investigate the organization of mobility programs at the degree course level. We investigate the number of destination countries students from a given degree course can choose from, the competition for mobility program grants (taking the number of students applying and places available for a degree course into account) and the level of the ability threshold students need to meet to be allocated a grant. We also consider the share of intake of disadvantaged students by degree course.

Research question 2: Are degree course characteristics, including ISM opportunities, competition, selection and intake of disadvantaged students, important for explaining the unequal uptake of Erasmus mobility?

3. Data and methods

We exploit rich administrative data from the oldest European university, the University of Bologna in Italy. Attracting approximately 5% of all students enrolled in higher education in Italy every year (Italian Ministry of Higher Education and Research 2022), the University of Bologna is the second largest public university in Italy and among the Italian universities with the strongest tradition of participation in the Erasmus program. Of all Bologna students graduating in 2019, 11.2% spent a period studying abroad with an Erasmus scholarship (or through another EU program), while this figure is 8.7%, on average, across the majority of all other Italian higher education institutions.³ Given Granato et al. 2024 (Table B1), the sample of students starting a study career at the University of Bologna is fairly representative of the population of Italian higher education students across some basic characteristics, including the share of female students and the distribution of students by field of study. Nevertheless, this data source (publicly available aggregated data released by the Italian Ministry of Education) does not allow investigating comparability of students starting a career at the University of Bologna and at other Italian universities in terms of characteristics like socioeconomic background and ability levels. Consequently, we cannot claim that students at the University of Bologna are representative of Italian higher education students in general.

We use administrative records on individuals’ demographic and study careers for students who enroll on all degree courses offered by the University of Bologna between the academic years 2010/11–2018/19. First, we match these records with Erasmus applications during the same time. These data contain a rich set of information on application steps, the destination country, the host institution, the student’s application score for all applications made and the threshold by mobility scheme. Second, we combine the University of Bologna administrative data with graduate cohort survey data from 2012 to 2019 deriving from the Graduates’ Profile Survey administered every year by the inter-university consortium AlmaLaurea. Upon graduation, students are required to answer a rich set of questions on their experience in higher education, their demographic characteristics (gender, country of birth, region of residence upon university enrollment), their secondary school education and family socioeconomic background (high school track attended and exit grade, mother and father education and last occupation).

For our analysis, we focus on bachelor students who graduated between 2012 and 2019 (approximately 46% of university entrants between 2010/2011–2018/2019) and who answered the AlmaLaurea survey (the yearly average response rate is very high with 93%). For more details on sample selection, see Table A1 in the Appendix. Our final sample comprises 46,096 students in 102 degree courses.

We construct binary variables indicating whether the student is female, foreign-born and moved to another region to attend university (the latter being used as a proxy for students’ attitudes towards mobility, Sorrenti 2017). To measure students’ academic ability, we build indicators of the student having attended an academic high school track versus a vocational track (even though the former track is more academic, both tracks allow full access to higher education in Italy) and of having exited high school with relatively high or low grades or in between (measured by being in the top or bottom 25th percentile of the distribution of high school final exam grade or in between in the final sample). We identify disadvantaged students as those having no parent with tertiary education. Furthermore, using information on households’ ‘social class’ we construct a rough proxy of likely financial difficulties, by identifying students whose both parents hold as highest professions blue-collar occupations.

Finally, thanks to the availability of university administrative data, which cover the entire population of students and – importantly – report the specific degree courses in which they are enrolled, we are able to measure degree course characteristics by aggregating from the population of students across degree courses. We calculate the total number of different mobility programs available for students (called ‘opportunities’ thereafter). In addition, we compute the weighted ratio of the single Erasmus scheme’s total number of Erasmus grants available over the total number of applications (‘competition’) as well as the weighted average of the single Erasmus scheme’s threshold score for qualifying for the Erasmus grant (‘selectivity’), where the weight is, for each degree course, the ratio of students applying for that single program over the total of students applying for Erasmus. Our degree course aggregates are calculated separately by year, across which we take the average.

Methodologically, we investigate research question 1 on the importance of selection stages for unequal Erasmus uptake by comparing how the share of disadvantaged students differs among the student populations at different selection steps. Regarding research question 2 and to investigate the unequal application by socioeconomic background (the choice of which will be explained later), we first run single-level logistic regressions not taking degree course factors into account.

Let $y_i$ denote the mobility of student i at Bologna University: $y_i=\{\begin{array}{c}1-\mathrm{student}\mathrm{applies}\mathrm{to}\mathrm{Erasmus}\mathrm{mobility}\\ 0-\mathrm{student}\mathrm{does}\mathrm{not}\mathrm{apply}\mathrm{to}\mathrm{Erasmus}\mathrm{mobility}.\end{array}$Then, the probability of students’ application by a logistic model for application

$p_i=\mathrm{Pr}(y_i=1)$ can be written as $\mathrm{logit}(p_i)=b^T{x}_i,$where $x_i$ is a vector of individual-level covariates and b is a vector of regression coefficients. Single-level logistic regression results provide an estimated association between socioeconomic background and mobility, unconditional and conditional on individual-level characteristics.

In a second step, we consider degree course characteristics by employing a multilevel approach. This allows us to estimate the variance partition coefficient (VPC), which provides the proportion of variation in the underlying student application propensity that is due to differences between degree courses. The multilevel model can be written as follows.

Let $y_{\mathrm{ij}}$ denote the application of student i in degree course j coded as $y_{\mathrm{ij}}=\{\begin{array}{c}1-\mathrm{student}\mathrm{applies}\mathrm{to}\mathrm{Erasmus}\mathrm{mobility}\\ 0-\mathrm{student}\mathrm{does}\mathrm{not}\mathrm{apply}\mathrm{to}\mathrm{Erasmus}\mathrm{mobility}.\end{array}$Then, the probability of student application by a general two-level random coefficients logistic model for application $p_{\mathrm{ij}}=\mathrm{Pr}(y_{\mathrm{ij}}=1)$ can be written as $\mathrm{logit}(p_{\mathrm{ij}})=b^T{x}_{\mathrm{ij}}+u_j^T{w}_{\mathrm{ij}},$where $x_{\mathrm{ij}}$ is a vector of student and degree course-level covariates and $w_{\mathrm{ij}}$ is a subset of student-level components of $x_{\mathrm{ij}}$ with random coefficients $u_j$ at the degree course level.

4. Results

Table 1 presents the descriptive statistics for both individual and degree course variables. On average, 16.8% of students apply for student mobility, 16.2% fulfill the requirements for the grant, 13.6% are awarded the grant and 11.2% take part in Erasmus. Hence, from the application to the uptake of the grant, 5.6 percentage points (or 33%) of original applicants drop out. 66% of students have both parents who do not hold a tertiary degree and 19% have parents who hold blue-collar occupations. On average across degree courses, we find that students have the choice of 35 mobility programs (opportunities), for 100 applicants there are about 76 Erasmus places available and the application threshold score for receiving a mobility offer is 62.

Research question 1: At which stage of the selection process do disadvantaged students drop out: at the application phase, the eligibility screening, the ability selection or the final uptake decision?

Figure 1 examines the share of students with lower socioeconomic backgrounds throughout the application process. The baseline ‘i’ shows that 66% of all Bologna graduate students have lower-educated parents. When looking at the population of applicants (‘ii’), only 55.4% of students have lower-educated parents, indicating that disadvantaged students are highly underrepresented among applicants. The share of disadvantaged students decreases only slightly during eligibility screening (‘iii’) and ability selection (‘iv’). The latter is surprising since, given that students with higher-educated parents tend to have on average higher ability, one could have assumed that the ability selection would result in a larger decrease in the disadvantaged students’ representation at this phase of the selection process. Disadvantaged students are slightly less likely to accept a grant if offered. Each selection step leads to a decline in the representation of students with lower-educated parents.

Figure 1. Percentage of students with lower educated parents in different population groups of the application process.

Table 1. Summary statistics.

	Full sample		Higher educated parents		Lower educated parents
Individual level variables:	Percentage	se	Percentage	se	Percentage	se
Probability of applying for Erasmus	16.8	0.17	22.1	0.33	14.1	0.02
Probability of qualifying for Erasmus grant	16.2	0.17	21.4	0.33	13.5	0.02
Probability of being awarded Erasmus grant	13.6	0.16	18.1	0.31	11.3	0.18
Probability of participating in Erasmus	11.2	0.15	15.0	0.29	9.2	0.17
Low parental education	66.0	0.22	0.0	0.00	1.0	0.00
Low Socioeconomic-status (SES)	18.9	0.18	3.7	0.15	26.7	0.25
Academic High School (HS)	75.8	0.20	88.1	0.26	69.5	0.26
Top 25% HS final grade	25.1	0.20	25.8	0.35	24.7	0.25
Bottom 25% HS final grade	26.0	0.20	25.2	0.35	26.4	0.25
Female	58.2	0.23	54.3	0.40	60.2	0.28
Moved region	39.4	0.23	41.6	0.39	38.2	0.28
Foreign born	4.3	0.09	4.4	0.16	4.3	0.12
Degree course variables:	Mean	SD	Mean	SD	Mean	SD
Erasmus opportunities	34.5	32.0	37.2	33.2	33.1	31.3
Erasmus selectivity	62.0	8.3	62.3	8.4	61.8	8.3
Erasmus competition	0.76	0.285	0.74	0.268	0.78	0.293
Share of low parental education students in percent	66.0	9.8	63.2	9.1	67.5	9.8

Note: Summary statistics computed on the final sample: 46,096 observations. See Table A2 in the Appendix for the definition of variables. 'se' denotes standard error and 'SD' standard deviation.

In sum, about 90% of the final socioeconomic gap in Erasmus uptake is due to the application process (66.0%−55.4%)/(66.0%−54.2%). For all fields of study examined separately, we consistently find that the largest loss of disadvantaged students appears when comparing the entire student population to the applicant population (from i to ii). Nevertheless, there are sizable differences by field of study. For example, students with lower educated parents are 64.1% of all students enrolled in natural science, ICT, mathematics and statistics but only 49% of all Erasmus applicants enrolled in this field, reflecting approximately a 15 percent point decline in disadvantaged students’ representation at the application phase. On the other hand, disadvantaged students represent 67.5% of all students enrolled and 64.1% of students applying for mobility for the field of study ‘languages’, showing a decline of just approximately 2.5 percentage points. (See Figure A1 in the Appendix for results by field of study.) Similarly, Schnepf and Colagrossi (2020) find a better representation of disadvantaged students among credit mobile students in language courses, assuming that this is due to student mobility being an integral part of the language study programs.

The application phase is also important for other related student characteristics (see Table A3 in the Appendix). 67% of ability segregation (focusing on upper secondary school results) takes place at the application phase. This indicates that students with lower academic ability feel more deterred from applying perhaps due to fear of not passing the ability selection process or application requirements being perceived as too high.

Given the striking result that most of the unequal uptake in Erasmus mobility derives from the application decision, we investigate why disadvantaged students shy away from applying to ISM. In line with the literature, we first consider individual-level characteristics.

Table 2 provides marginal effects for both logistic and multilevel logistic regression models. Model a shows that students with lower parental education are about 8 percentage points less likely to apply. This is our baseline difference and in line with descriptive statistics of Table 1, showing that about 22% of students with higher but only 14% of students with lower-educated parents are applying for an Erasmus grant. Conditioning on lower parental occupation (a variable equal to 1 if parents have blue-collar occupations) decreases the application gap by just 0.5 percentage points (Model b, from 7.7 to 7.2 percentage points). However, it is important to note that students whose parents have lower level occupations, hence likely lacking financial resources, are, conditional on parental education, 2 percentage points less likely to apply.

Table 2. Logistic and multilevel models for probability of applying for Erasmus: marginal effects.

	Logistic models				Multilevel models
	Model a	Model b	Model c	Model d	Model e	Model f	Model g	Model h
Individual level characteristics:
Low parental education = 1	−0.077***	−0.072***	−0.059***	−0.061***		−0.046***	−0.052***	−0.054***
	(0.004)	(0.004)	(0.004)	(0.004)		(0.004)	(0.004)	(0.004)
Low SES = 1		−0.022***	−0.017***	−0.021***		−0.014***	−0.016***	−0.017***
		(0.005)	(0.005)	(0.005)		(0.004)	(0.004)	(0.005)
Top 25% HS final grade = 1			0.030***	0.023***		0.011***	0.013***	0.013***
			(0.004)	(0.004)		(0.004)	(0.004)	(0.004)
Bottom 25% HS final grade = 1			−0.030***	−0.026***		−0.013***	−0.014***	−0.014***
			(0.004)	(0.004)		(0.004)	(0.004)	(0.004)
Academic HS = 1			0.065***	0.049***		0.028***	0.031***	0.032***
			(0.004)	(0.004)		(0.004)	(0.004)	(0.005)
Female = 1				0.040***		0.004	0.004	0.005
				(0.003)		(0.003)	(0.004)	(0.004)
Moved region = 1				0.079***		0.037***	0.041***	0.043***
				(0.004)		(0.004)	(0.004)	(0.004)
Foreign born = 1				0.027***		−0.000	−0.000	−0.000
				(0.009)		(0.007)	(0.008)	(0.008)
Fields of study:
Languages							0.316***	0.243***
							(0.094)	(0.082)
Education							−0.151***	−0.066
							(0.044)	(0.061)
Arts and humanities							−0.084**	−0.064**
							(0.037)	(0.032)
Natural sciences, ICT, maths and stats							−0.153***	−0.113***
							(0.026)	(0.025)
Engineering, manufacturing and constructions							−0.180***	−0.166***
							(0.027)	(0.023)
Health and welfare							−0.156***	−0.098***
							(0.029)	(0.031)
Degree course variables:
Share low parental educ. students								−0.354***
								(0.081)
Erasmus opportunities								0.040***
								(0.016)
Erasmus selectivity								0.002
								(0.005)
Erasmus competition								−0.007
								(0.009)
Constant					−2.256***
Sigma u					1.303***	1.206***	0.525***	0.334***
					(0.206)	(0.192)	(0.091)	(0.066)
VPC					0.284	0.268	0.138	0.097

Note: standard errors in parentheses. All models include enrollment cohort and graduation cohort fixed effects.* p < 0.10, ** p < 0.05, *** p < 0.01. In models g and h the omitted category for field of study is ‘Social sciences, business and law’. Variables are defined as in Table 1. In model h, the variables Erasmus opportunities, Erasmus selectivity, Erasmus competition are standardized to have mean 0 and sd 1. The number of observations for all models is 46.096 students.

Comparing the entire student population with the final Erasmus population, we found that two-thirds of the change in the ability levels between populations is due to the application process (table A3 row 1 in the Appendix), possibly indicating that lower ability students are deterred from applying since they fear the ability selection. Model c confirms this. Indeed, those students in the lowest performance quartile of upper secondary school exams are 3 percentage points less likely to apply. In addition, students who previously enrolled in the more demanding academic Italian secondary tracks are as much as 6.5 percentage points more likely to apply. Since ability is correlated with socioeconomic status (see Table 1), we find that conditioning on ability decreases the socioeconomic gap in applications (comparison Model a with Model c). However, it can only explain about 1.3 percentage points of the 7.7 percentage point unconditional socioeconomic gap (so less than 20%) in applications. Thus, while the ability selection process might deter some disadvantaged, lower-ability students from applying, this is not the driving force of the socioeconomic gap among applicants. Consequently, if ability selection into Erasmus mobility would be discontinued, the underrepresentation of disadvantaged students among applicants would probably only slightly decrease.

Conditioning on gender and proxies of student affinity to moving (a binary variable coded 1 if the student moved region (Italian NUTS 3) to study in Bologna and an indicator for being born abroad) does not decrease the socioeconomic gap in application (Model d). Consequently, previous mobility experiences of students cannot be linked directly to the socioeconomic gap in applying to Erasmus (conditional on ability). Nevertheless, attitudes towards mobility are important for application probability. In line with the literature (Wiers-Jenssen 2011), students who moved to study within Italy are as many as 8 percentage points and those who are born abroad are still 3 percentage points more likely to apply for Erasmus.

Even if we condition on proxy variables for ability, parental occupation, gender and mobility affinity, the socioeconomic gap for applications remains still as high as 6 percentage points (Model d). This leads to the question of whether other, especially meso-level characteristics of mobility organization, could affect application levels and explain the socio-economic gap in applications.

Are degree course characteristics, including opportunities, competition, selection and intake of disadvantaged students, important for explaining the unequal uptake of Erasmus mobility?

To investigate our second research question, we switch from logistic to multi-level models. This allows us to examine whether the association of student-level characteristics such as socioeconomic background varies across degree courses (random coefficient models).

Model e provides the null model showing the Variance Partition Coefficient (VPC) independent of any control variables. As much as 28 percent of the variation in student applications can be explained by differences between degree courses. Consequently, degree course programs are very important for understanding Erasmus applications. Figure A2 in the Appendix plots the degree course effects obtained from the null model for each of the 102 degree courses in our sample, shown in rank order together with 95% confidence intervals. Certain degree courses are characterized by a significantly lower probability of applying for Erasmus than the average, for example, some engineering courses (for which the Erasmus application rates are between 1 and 2%). Degree courses with significantly higher Erasmus application rates are in the fields of economics and language, with some courses having up to 85% of students applying for Erasmus.

Multi-level model f controls for the same individual level variables as the simple logistic regression Model d, but takes degree course random effects into account. Indeed, the socioeconomic gap in Erasmus applications shrinks from 6.1 (Model d) to 4.6 percentage points (Model f), a decline of 1.5 percentage points. Consequently, how students are distributed between degree courses has an impact on the socioeconomic gap in applications. This is due to the fact that students with lower socioeconomic backgrounds tend to enroll in degree courses that have lower overall Erasmus applications. The share of students with low socioeconomic backgrounds and the average mobility application rate by degree course is correlated with r = −0.45 (see Figure A3 in the Appendix).

Moreover, the size of all other individual level coefficients shrinks by about one-half and gender is not any more important for predicting Erasmus application. We find similar results if we use Erasmus participation (which was called (v) previously in Figure 1) in contrast to application (ii) as dependent variable (see Table A4 in the Appendix). This indicates that models not taking the clustering of students in degree course levels into account provide estimates for individual level factors that derive from both the actual importance of the individual-level factor for Erasmus participation as well as its clustering across degree courses with different shares of Erasmus participation.

Model g also conditions on fields of study, thereby taking into account that degree courses of specific fields of study might differ on average in Erasmus uptake. Indeed, the coefficients are significant, but do not lead to any significant changes in the coefficient of students with lower socioeconomic backgrounds (comparison between Model f and g). Nevertheless, conditioning on the field of study reduces the VPC by only approximately one-half, from 0.27 (Model f) to 0.14 (Model g). The remaining VPC of 0.14 implies that fields of study is too general an aggregation level for explaining variation in applications (and participation), since degree course level matter beyond fields of study.

Students clustering in degree courses could be important for explaining Erasmus applications due to the socioeconomic composition within the degree courses. We discussed that degree courses with a higher share of applicants on average are also those with a lower share of disadvantaged students (Figure A3 in the Appendix). Consequently, we investigate the conditional ‘effect’ of socioeconomic background composition in degree courses in model h (besides other variables describing the degree course level organization). An average degree course is attended by 66 percent of students with lower educated parents with a standard deviation of about 10 percentage points (see Table 1). In our model, the variable ranges from 0 to 1 measuring proportions. The share of disadvantaged students in a degree course is greatly correlated with application rates in general even conditional on all individual and degree course variables. Being enrolled in a degree course with a proportion of disadvantaged students one standard deviation higher than average (e.g. moving from the mean of 0.66 to 0.76) is associated with a lower probability of application by 3.5 percentage points (one standard deviation times 0.354). At the same time, conditioning on this course level variable does not decrease the size of the coefficient capturing parental education, showing that this association is ‘hidden’ in the random effects of the degree course variable.

How important are other variables capturing the management of Erasmus mobility at the degree course level? An increase in the degree courses’ total number of different mobility programs available for students (variable ‘Erasmus opportunities’) of one standard deviation is associated with an increase in the probability of application for a student of 4 percentage points (the variable is standardized in our model). Table A4 in the Appendix provides results for the same model but with Erasmus participation (instead of application) as dependent variable and finds the same relationship. This strong correlation could reflect that degree courses offering more destinations attract more students in terms of applications and participation.⁴

Instead, the extent of selectivity, which is the stringency of the threshold below which applicants are rejected, and of competition measured by the ratio of the total number of Erasmus grants available over the total number of applications are not significantly correlated with the student’s probability of applying for Erasmus. This could be due to applicants not knowing how degree courses differ in terms of the toughness of the selection process and of the competition for Erasmus grant. However, the same applies once Erasmus participation is concerned (see Table A4 in the Appendix), which would indicate that, beyond individual ability already conditioned for in the model, the degree course selectivity (itself based on ability) and competitions levels do not further contribute to Erasmus participation. Consequently, our results imply that the differently tough ability selection at the degree course level does not influence significantly Erasmus mobility and the socioeconomic gap in mobility uptake. Combined with the previous result showing that conditioning on ability can only slightly reduce the socioeconomic gap in applications, this indicates that the ability selection process is unlikely to greatly contribute to the socioeconomic gap in mobility uptake.

Most importantly, conditioning on degree course level variables measuring management issues of Erasmus mobility and student composition within degree courses decreases the VPC from 0.14 to 0.10. We cannot explain ‘away’ the remaining 0.10 of the VPC, indicating that there might be additional degree course variables of importance we do not have information on (e.g. different support for students applying by degree course etc.).

To investigating whether results are robust to the weighting procedure used to create these degree course variables, we run the same regressions but using different possibilities for degree course variable creation. Results are mainly very similar (see Appendix A5 for more details on the robustness checks).

5. Conclusions

Literature trying to explain the socioeconomic gap in ISM mobility generally focuses on individual-level characteristics, like students’ ability. In contrast, this study investigates whether ISM program organization fosters disadvantaged students’ lower participation in ISM. The paper examines Erasmus, the biggest ISM policy in place for European countries. It offers grants to students for spending a part of their studies abroad while remaining enrolled in their home country’s tertiary education institute. We exploit rich administrative data from the second largest public university in Italy, the University of Bologna, merged with survey data covering 46,096 students in 102 degree programs during the years 2010–2019.

The first key finding of the study is that, focusing on selection into the study-abroad program, 90% of the socioeconomic inequalities in Erasmus participation derive from the first selection stage of students: the application. Students with lower-educated parents are considerably less likely to apply. All other selection stages (eligibility checks, ability selection and final participation decision) discriminate against disadvantaged students, too, but to a much lesser amount. The drop in the representation of disadvantaged students among applicants compared to all graduates is consistent across all fields of study, but is especially large for natural sciences, ICT, mathematics and statistics while very moderate for languages. This result is in line with Van Mol and Perez-Encinas (2022): having ISM as an integral part of a degree course is likely to reduce the socioeconomic gap either since the application is considered as ‘normal’ or because there is no selection process.

Given that most of the socioeconomic gap in Erasmus participation is grounded on disadvantaged students not applying for Erasmus, the study then examines in detail the determinants of Erasmus application and how they link to socioeconomic status. For doing so, the study examines the role of individual-level variables, including students’ ability and previous mobility, and, in addition – to the knowledge of the authors for the first time – the importance of ISM organization for unequal application measured at the disaggregated level of degree courses. The second key finding of the study is that, after having accounted for students’ characteristics, conditioning on students clustering in degree courses with random effects decreases the socioeconomic gap in applications by about 1.5 percentage points. This indicates that students with lower socioeconomic backgrounds tend to enroll in degree courses that have lower overall Erasmus applications. The degree course characteristics that we are able to measure – including their socioeconomic composition and measures of the Erasmus program management – explain more than 70% of the VPC of the null model. However, they cannot explain the socioeconomic gap in applications further. Overall, degree course characteristics are at least as important for explaining the socioeconomic gap in uptake as individual-level characteristics. For the latter only ability matters: conditioning on ability decreases the socioeconomic gap in applications by about 1 percentage point. This decrease however reflects less than 20% of the socioeconomic gap in applications.

Taken together, these results indicate, first, that higher education institutions should focus on supporting disadvantaged students during the application phase, including addressing any misconceptions about the challenges of the selection process that may discourage disadvantaged lower-ability students from applying for international mobility. Additionally, these interventions should be tailored and implemented at the degree course level, where mobility programs are managed and students are most closely engaged, focusing on fostering incentives for mobility in those degree courses that are attended by the less privileged.

One important implication of our paper is that, given the finding that as much as 28 percent of the variation in Erasmus applications derives from differences between degree courses, any study examining Erasmus mobility would benefit from taking degree courses into account. However, degree-course level data are only accessible with administrative data sets, which are rather uncommon for tertiary graduates. In fact, this study uses rich administrative data from only one single university in Italy. This clearly limits the external validity of the analyses. However, only this data set allows measuring meso-level factors relevant for students’ mobility decisions largely unexplored in previous literature because of limitations of data covering many universities.

Moreover, even though the paper exploited rich administrative data, regression results indicate that only 25% of the socioeconomic gap in applications can be explained. This is likely to be due to additional individual and degree course variables not covered in the data set. Focusing on the individual level, future research would benefit from data that include measures of students’ personality traits and better information on their financial situation. The latter could have an even greater influence in the most recent times, characterized by higher economic uncertainty following Covid-19 and the energy crisis. Moreover, the Covid-19 pandemic – besides the obvious short-term negative impact on students’ mobility – might have changed how students perceive studying abroad. The replacement of physical mobility with ‘virtual mobility’ could allow reaching more students (European Commission 2023), especially the more financially constrained ones.⁵ At the degree course level, it is likely that information flow and support on ISM applications differ by degree courses. Therefore, more in-depth information on ISM organization by degree course level would be needed to fully understand how the clustering of students in degree courses links to the socioeconomic gap in applications.

Disclaimer

The views expressed are purely those of the authors and may not under any circumstances be regarded as stating an official position of the European Commission. The University of Bologna does not accept responsibility for any inferences or conclusions derived by third parties from data or other information supplied by the University of Bologna.

Acknowledgements

We thank conference participants of the ‘2022 Conference of Cost Action CA20115, European Network on International Student Mobility: Connecting Research and Practice (ENIS)’ (October 2022), the Higher Education and Equality of Opportunities Workshop (June 2023) and the 38th Italian Association of Labor Economics (AIEL) annual conference (September 2023) for very helpful comments and suggestions. Our special thanks go to the office for planning and data analysis and the office for Erasmus + agreements and mobility, especially Danilo Cinti, Carmela Tanzillo and Camilla Valentini, for making the data available for this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 International student ‘credit’ mobility is different from ‘degree’ mobility, where students enroll in and graduate from an entire study program abroad. The Erasmus program was established in 1987 to enable European university students to spend a study period abroad in an EU member state, supported by mobility grants. In 2014, the name of the program changed from Erasmus to Erasmus+ to indicate its expansion to a wider array of activities (including vocational training, school, adult education and sport).

2 These meso-level factors of the sending university relate partially also to the characteristics of destination institutions, which provide further pull factors potentially playing a role in students’ mobility decisions. To the best of our knowledge, there is no recent research exploring this aspect.

3 Source: AlmaLaurea 2019. Italy is one of the leading EU countries in terms of Erasmus participation. In 2019, Italian universities accounted for approximately 14% of all credit mobile graduates in Europe (Eurostat 2024). Unfortunately, we cannot provide the same figures for Erasmus applications, given the unavailability of data on Erasmus applications, even aggregated, covering the entire population of higher education students either in Italy or in any other European country.

4 Our variable ‘Erasmus opportunities’ gives a measure of the heterogeneous offer of mobility programs across degree courses that is likely to be the result of exogenous supply factors as well as the endogenous response of the supply to students’ demand for international mobility.

5 The evidence on the impact of Covid-19 on international student mobility is still limited. The most recent figures document that the share of mobile students has not significantly decreased in the years after the pandemic across OECD countries (OECD 2023, Table B6.1), while credit mobility among European graduates has slightly declined (European Commission 2023, Figure 31).