Pattern detection in human-led Refugee Status Determination practice

Asylum law is a legal field where administrative routine meets potential life and death decisions. Decision makers must effectively enforce national sovereignty by ensuring both that no individual is returned to a place where they risk persecution, and that international protection only is granted to those with genuine protection needs. While the consequences of a false negative in this context can be fatal, the reality of the decision-making process is often described as routinised, biased and rooted in a culture of distrust.1 At the same time, refugee status determination (RSD) is a field where ‘legal quickening’ is a political ambition in many jurisdictions,2 and several countries, including Denmark, are at various stages of testing and implementing new technologies in this domain.3 Both the recently agreed EU Migration and Asylum Pact as well as extensive funding to border technologies implies encouragement to use technology powered tools to fasten the process of distinguishing between those with legitimate and illegitimate protection needs. 

To better understanding of the data making up potential training data for recommendation systems in the RSD realm this dissertation is concerned with mapping patterns of human decision making in asylum cases. It makes three contributions: Substantively it examines the patterns of human decision-making in asylum appeal cases to understand the practice and how it has developed over time. Second, it makes a methodical contribution to the field of computational legal studies. Previous scholarship has to a large extend analysed large datasets to revel descriptive patterns based on meta data or has left it to qualitative studies to analyse selected cases to explain how and why decisions were made. By combining large-N “deep data” with statistical modelling and experimental methods, this project seeks to move beyond descriptive patterns and quantitatively identify the factors underlying decision making. The third contribution lies in the combination of the two previous: By using the same statistical models upon which prediction and recommendation models are built, the dissertation treats this written documentation of practice as if it were training data. Where most qualitative and doctrinal studies of asylum decisions are based on the decision writing of adjudicators, I seek to leverage methods from data- and social sciences to move beyond the reasoning of decisions and identify the latent patterns in the aggregated and historic practice making up what algorithms would seek to mimic. This perspective opens up a new way of seeing the hidden patterns and statistical biases, potentially causing indirect forms of discrimination, that have so far remained below the surface.