Who offers English-medium instruction? Exploring university characteristics with random forests
Journal article, 2025
English-medium instruction (EMI) has spread throughout European higher education since the turn of the century. Whilst much valuable research has been conducted on pedagogic concerns, the growth of EMI across countries, and macro-level drivers, we know surprisingly little about the kinds of institutions that offer EMI. This paper fills this gap by exploring the meso-level (institutional) predictors of EMI in European higher education institutions using interpretative machine learning. Random forests are employed to analyse data from the European Tertiary Education Register and Study Portals, focusing on meso-level features linked to EMI adoption. The model achieves an accuracy of 84% and a Cohen’s Kappa of 0.64, indicating the strong predictive performance of meso-level features. The most important feature was PhD discipline diversity, suggesting that discipline-diverse institutions were more likely to offer EMI than specialist institutions. An exploration of all features suggests that the archetypal EMI-offering institution in Europe is disciplinarily diverse, large, internationally oriented, research intensive, resource rich, and postgraduate focused. These findings are discussed through the lens of institutional isomorphism and the global spread of EMI, with divergence in the case of EMI potentially giving rise to elitism or openness, and the implications this has for higher education planning and policy.
machine learning
higher education
language policy
English-medium instruction
random forests