'Thou shalt not kill' a democide study

Abstract

This Master Thesis analyzes the factors that influence the occurrence of democide. The study focuses on a set of fifty three developing countries from 1960 to 2001. The dependent variable is democide and independent variables include: democide, International Monetary Fund (IMF) credits, consumer price index (CPI), gross domestic product (GDP) per capita, Gini index, military expenditures as a percentage of GDP, incidence of intrastate conflict, and democracy. An important limitation of the current literature has been the shortage of studies using yearly democide dataset. Even fewer of those studies strive to incorporate the actual number of people killed; instead they use democide as a dummy variable in their models. This study aims to fill this gap. The objective of this thesis is to answer an important question of whether the economic and social factors such as gross domestic product, IMF credits, income inequality, inflation, conflict history, and democracy play an important role in influencing murders by governments in the less developed and developing countries. The results when using democide as a dummy variable are consistent with the previous studies showing that conflict history and GDP per capita increase the likelihood of democides. This study also reveals that an increase in IMF funding last year can potentially lead to an increase in democide this year. Controlling for democracy does not have a significant effect on the regressions. The history of conflict remains statistically significant at one percent significance level in both fixed and random effect models using democide variable as an actual number of people killed regardless whether democracy variable is included or not. Military expenditures as a percentage of GDP also appear to be statistically significant at ten percent level. Even when IMF variable is replaced by IMF as a percentage of GDP and GDP per capita variable is replaced by a squared GDP, the history of conflict variables remain consistently statistically significant at one percent significance level in logit, probit, fixed, and random effect models.