Bad neighbourhoods, bad influences: The role of external factors in predicting genocide

Too often, poor accuracy limits the performance of risk assessment models for genocide. This casts doubt on new predictions, fomenting inaction amongst the international community and missed opportunities for timely interventions. These models focus almost exclusively on internal country variables as predictors. Genocide studies offers almost no published research examining the spread of genocide across borders. Yet, a global examination of genocide onsets between 1955 and 2016 suggests a geographical clustering effect. This prompts the question of whether this is due to similar internal factors, shared by countries within a geographic area, or if external factors play a greater role in genocide than previously assumed.

The key expectation of this thesis is that certain external factors do affect a country’s risk of genocide. Hence, the research investigates what these external factors are and whether their inclusion in risk assessment models improves their predictive performance. It does this by taking an empirically based, quantitative approach. Logistic regressions are used as a proof-of-concept; establishing statistical significant effects of specific external factors. The thesis then pivots towards predictive modelling and compares the performances of machine learning classification models, both with and without external factors.

The findings suggest that external factors do matter. Building on democratic peace and Deutschian Integration theory, the strength of democracy in a country’s immediate neighbourhood proves to be a robust indicator of a country’s risk of genocide. Similarly, a country with strongly democratic trading partners appears less at risk. In contrast, a prior instance of genocide within a country’s immediate neighbourhood serves to increase a country’s risk. Ultimately, these results recommend that external factors be included in future risk models for genocide.