Non-compliance of the European Court of Human Rights decisions: A machine learning analysis

Abstract

The paper investigates all (971) non-executed pending leading cases of the European Court of Human Rights (ECtHR) between 2012 and 2020 through Machine Learning (ML) techniques. Drawing on the extant scholarship, our interest on compliance has centred on state level and case level variables. For the identification of important variables, four databases have been used. Each country party to the European Convention on Human Rights (ECHR) received 232 distinct factors for eight years. Since we aim to make a parameter estimation for a high dimensional data set, Simulated Annealing (SA) is employed as feature selection method. In the state level analysis, Support Vector Regression (SVR) model has been applied yielding the coefficients of the variables, which have been found to be important in spelling out non-compliance with the ECtHR decisions. For the case level analysis, different cluster techniques have been utilized and the countries have been grouped into four different clusters. We have found that the states that have relatively high levels of equality before the law, protection of individual liberties, social class equality with regard to enjoying civil liberties, access to justice and free and autonomous election management arrangements, are less susceptible to non-compliance of the decisions of the ECtHR. For the case level analysis, type of violated rights, the existence of dissent in the decision and dissenting votes of national judges for their appointing states affect the compliance behaviour of the states. In addition, a notable result of the research is that if a national judge casts a dissenting vote against the violation judgment of the ECtHR involving the state that appointed him/her, the judgment is likely not to be executed by the respondent state.