The dataset svac_geocoded contains select variables from SVAC version 1.0 by Dara Kay Cohen and Ragnhild Nordås (2014), and is filtered to observations of rebel actors (coded as actor_type 3 in SVAC) in Africa where sexual violence was recorded in the ai_prev, hrw_prev or state_prev variables. It also contains several new variables coded by the author using the SVAC variables. These variables are: location_specific, sv_location, sv_nonspecific, and sv_dummy. The variable SV_dummy is the aggregated prevalence measure by which these observations were filtered, assuming a value of 1 if any of the aforementioned prevalence measures records any level of sexual violence for a rebel actor - year. It has a value of 1 for all observations in the dataset. The variable location_specific assumes a value of 1 if the variable location_text for an observation in SVAC 1.0 specifies one or several locations in which sexual violence occurred that could be conntected to one or several first order admin units. All identifiable admin units are listed and each original observation was duplicated once for every admin unit. The variable sv_location contains the admin unit names. Accordingly, the dataset has a admin unit - year structure, unlike SVAC 1.0 which has an actor - year structure. The variable sv_nonspecific assumes a value of 1 if there were locations in the location_text variable that could not be tied to a specific admin unit.
Steps to reproduce
To use SVAC Geocoded, load the dataset into a data management software and either merge with a dataset with the same structure or restructure the geocoded data to merge with an actor - year dataset. If the user wishes to prepare the data for replicating the author's analysis,¹ the suggested approach is to aggregate the data into an actor - year structure and list each unique admin unit in the sv_location variable in a single variable. Then, separate that variable into a sufficiently large number of new variables (e.g. sv_location_1, sv_location_2, etc.), each containing one first order admin unit in which sexual violence occurred. Then merge the geocoded dataset with SVAC 1.0 by actor and year. Finally, merge this data with spatial data at the first order admin level and match the sexual violence location variables with these admin units. ¹ The analysis in Wieselgren, Herman. 2022. "Sexual Violence along Ethnic Lines? Revisiting rebel-civilian ethnic ties and wartime sexual violence".