Who is in charge? Explaining the European Union’s changing implementation design through a machine-learning application to the syntactic structure of EU laws

Over the past thirty years, the implementation of European Union (EU) laws has become increasingly complex as far as the actors participating in the policy process are concerned. On the one side, EU legislators have created a plethora of new supranational agencies with specific implementation prerogatives, both at the national and supranational levels. On the other side, they have increasingly relied on specific competent authorities for implementation at the national level. How can we explain these choices? Why do legislators rely on such agencies and competent authorities? Which benefits, risks, and incentives are associated with these decisions? In this paper, we subject to empirical corroboration a set of expectations that flesh out the rationale behind these choices. We also develop an innovative measure of implementation design that is valid and easily scalable to large amounts of data and represents a clear advantage over manual coding. The measure applies machine learning to the syntactic structure of more than 9,000 secondary EU laws adopted between 1958 and 2019.