APPENDIX MATERIAL FOR INTRODUCTION

Table A.1 shows that the subset of multiparty civil wars in the Fearon and Laitin (FL) (2003) dataset (by my definition of three or more internal warring groups) is relatively representative of my full set of fifty-three multiparty wars (which spanned a greater time period than the FL dataset and had slightly different rules for the number of fatalities). Specifically, I compared all available covariates present in both datasets for these two sets of wars. Despite scope and definitional differences, the fifty-three cases I identified did not differ significantly from the subset of thirty-six multiparty wars matching those in the FL dataset on any of the covariates tested, including duration, mortality rates (and log transformations of these two measures), or year of onset (Table A.1, compare column 1 to 2).

Comparing multiparty conflicts to non-multiparty conflicts within the FL dataset (Table A.1, compare column 2 to 3), we see that the subtypes of war seem to differ systematically on a number of covariates. In bivariate tests, multiparty conflicts were associated with significantly longer mean durations by a stunning eight years (p = 0.001). Similarly, when fitting a survival model using a Weibull distribution, multiparty wars were found to be 2.2 times longer (p < 0.000). Multiparty conflicts also had higher fatality rates when measured in natural logs (p = 0.0005), later years of onset (p = 0.024, reflecting the increasing rate of multiparty wars over the decades), higher ethnic fractionalization (p = 0.0009), a greater share of the population belonging to the largest ethnic group (p = 0.0003), and what may be a greater likelihood of employing symmetrical nonconventional war (SNC) (20% for multiparty vs. 6% for biparty, p = 0.10) as coded by Kalyvas and Balcells (2010b).