How do European higher education institutions internationalize?

ABSTRACT

This article explores how higher education institutions (HEIs) internationalize, employing information on the internationalization activities (IA), context and organizational characteristics of 431 HEIs from 33 European countries. A latent cluster analysis identifies three distinct clusters of HEIs with distinct portfolios of IA: basic, academic and entrepreneurial. The basic portfolio includes the most common IA, whereas IA requiring larger organizational capacity are rare. The entrepreneurial portfolio distinguishes from the academic portfolio as it also includes IA aimed to attract resources. We explore what contextual and organizational traits characterize HEIs with different IA portfolios. Small HEIs tend to display a basic portfolio, without national variations. On the contrary, strong national variations exist in the frequency of academic and entrepreneurial portfolios, which strongly relate to the actual and potential importance of tuition fees as a source of revenues.

KEYWORDS:

• Internationalization
• Internationalization activities
• student tuition fees
• configurational approach
• clusters and portfolios

Acknowledgements

We are grateful to the reviewers for their insightful comments. We thank Eva Egron-Polak and Ross Hudson of the International Association of Universities (IAU) for making available the data of the internationalization survey, Martina Vukasovic, Melissa Laufer, Jeroen Huisman and participants to the 30th CHER conference for their constructive comments, and Michael Wise for carefully proofreading the manuscript. We warmly thank Dominik Antonowicz, Tehri Nokkala, Attila Pausits and Javier Vidal for their help in determining data on tuition fees for Poland, Finland, Austria and Spain.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Marco Seeber http://orcid.org/0000-0002-0162-6289

Michele Meoli http://orcid.org/0000-0002-9438-0782

Mattia Cattaneo http://orcid.org/0000-0003-4400-089X

Notes

1 We are aware that IA may be developed also with reference to the so-called third mission of universities, such as in the case of international collaborations for the development of spin-offs. Given the scarcity of activity in this direction, and due to the fact that the IAU database employed in our analysis does not cover them, we limit our discussion to IA in teaching and research.

2 The survey also included the question ‘of the activities that are undertaken, which are given the highest priority’. The existence of such a question prevents ambiguity in the interpretation given by the rectors to the first question: this is clearly not addressing priority but the IA developed.

3 For an extended description of the rationales, see Seeber et al. (2016).

4 Double responses are attributed to the same cluster in 60% of the cases, compared to a random expected co-clustering of 36%. The difference in actual and random co-clustering is highly significant (p-value: .009**) (see Section 3.2).

5 The data have a three-level structure: 459 respondents nested into 431 universities, nested within 33 countries. Since most HEIs have only a single respondent, then respondents are selected at level 1 units. We employ a Markov Chain Monte Carlo (MCMC) method of estimation, since with a low number of level 2 units it provides more stable parameter estimates than the maximum-likelihood method of estimation (Stegmueller 2013).

6 The relationship between two variables may be very different when considering relationships within a group and between-groups. For example, cardiovascular death rates (Y) are higher in richer countries (income=X), but within each country it is poorer people who tend to be more at risk. Hence, the true relationship between Y and X is revealed only when both relations are considered jointly (Snijders and Bosker 2012, 29).

7 The fit statistics for 2, 3, 4 and 5 classes LCA are maximum log-likelihood: −2615; −2511; −2474; −2455; AIC: 5281; 5099; 5050; 5037; BIC: 5384; 5255; 5261; 5302. BIC of the 4-class solutions is similar to the 3-class solution, but the additional complexity does not seem to be worthwhile as it re-creates the academic and basic clusters, while splitting the entrepreneurial cluster into two clusters that are very similar except for off-shore campus, distant learning and development project.

8 We computed the Variance Inflation Factors (VIF) as a standard test for multicollinearity (values should be below 10). The VIF values are all below 1.17, which exclude multicollinearity among the predicting variables.

9 In binomial multilevel regression the proportion of variance at country level (VPC, variance partitioning coefficient) is computed as ơ²/(ơ² + 3.29); where ơ² = variance at country level (Snijders and Bosker 2012).

10 For some countries the number of HEIs is too small to achieve significant differences, such as for Denmark, Czech Republic and Ireland.

11 Public expenditure on education per tertiary education student (FTE).

12 Several jurisdictional bodies have intervened in 2017 against the decision of the Polytechnic of Milan to provide some degree courses only in English and oblige to always provide the same course also in Italian (e.g. Corte Costituzionale 2017).

Additional information

Funding

This work was supported by Fonds Wetenschappelijk Onderzoek [G.OC42.13N].