that had identical ID codes, incompatibility, location, main actors, and were separated
Applying GIS to Generate Spatial Variables
In order to generate data that are able to incorporate spatial attributes, we relied on
desktop geographical information systems (GIS). A GIS tool, such as ArcView or
ArcInfo, combines digital maps with relational databases that have statistical
functions, and enables the researcher to map and analyze variables that have a spatial
dimension. Several of our variables were generated this way, mapping the distribution
of specific regressors and intersecting the resulting theme layers with the conflict
map. In order to pursue this task we first needed spatial information on the relevant
variables. For the geographic distribution of conflicts, we mapped the units in the
Armed Conflicts in accordance with the latitude, longitude, and radius variables (c.f.
Buhaug & Gates 2002 or Gleditsch et al. 2002).
Dependent Variable: Duration of Internal Conflicts
The dependent variable is the duration of internal conflicts. Although the Armed
Conflicts dataset includes start-dates of the conflicts, it does not contain detailed
information on the termination of conflicts. Consequently, we used the Gates &
Strand (2005) dataset, which more precisely dates war start and termination. This
dataset offers a number of advantages. First as with all data that are part of the Armed
8
Deciding where to place the cut-off point between ongoing and new conflicts is not trivial and may
have a substantial impact on the results. Cut-offs less than 24 months seem to be more problematic than
longer cut-offs. Moreover the 24 month criterion possesses a certain prima facie validity. See Gates &
Strand (2005) for a more detailed discussion of coding civil wars involving intermittent fighting.
19
Convention