Initial descriptive analysis of conflict in South Sudan. We focus in particular on three types of events from the ACLED dataset:

Time span: January to June 2017, weekly level

Spatial visualization of conflict

Labeling of S. Sudan map.

All conflict from January to June 2017.

All conflict from January to June 2017

this is a time lapsed spatial visualization of the events of interest (EOIs) — battles, violence against civilians, and riots/protests — in South Sudan. Each of the EOIs is distinguished by color and shape on the map. To visualize potential evidence of lagged spatial autocorrelation, I include the last period’s EOIs with a higher transparency setting.

EOIs by state

Count of EOIs across the 32 states of South Sudan from Jan. to June 2017. Here we find that the number of riots/protests are in general quite small and strangely they only really take place in a handful of states. Violence against civilians is more widespread and we see that some states are notably more dangerous for civilians than others. These set of findings just speak to needing to account for variability in EOIs across states through some random or fixed effects structure for the states.

Events by state

Time series of EOIs

Time series visualization of each EOI for all 32 states. At the weekly level we see that in most states nothing is going on which is not too surprising. But even in those states that are more active the pattern can fluctuate from week to week, for example, in Yei River State the number of battles goes from zero in one week to six in the next and then back to zero. Based on a simple correlation analysis, I don’t think that we’ll be able to account for this type of variability by using the EOIs to predict one another. For example, across our time series of 32 states incidences of battles are correlated with violence against civilians at a 0.12 level and 0.004 for riots/protests.

Battle time series

Violence against civilians time series

Riots/protests time series