Approaches Towards the Identification of Patterns in Violent Events, Baghdad, Iraq

摘要

This work analyzed the spatial distribution of violent events as constructed from a content analysis of open source news reports. Data on 112 variables was available for 45 neighborhoods. The small sample limited the analysis to those 32 variables with at least four observations. The statistical analysis was done both for the original measures of event counts by neighborhood, and for binary variables that indicated the presence of events. Test statistics for spatial autocorrelation were computed for global patterns and local patterns, including global and local Moran's I, Geary's c, Moran Scatterplot, join count statistics, and local join counts. There was little evidence of systematic spatial structure at the neighborhood scale. Only for a variable indicating internal between hayy migration was there consistent indication of positive spatial autocorrelation, or clustering. Several other variables showed significant negative spatial autocorrelation at the local scale, suggesting that neighborhoods where violent events occurred are surrounded by neighborhoods without violent events. A few neighborhoods were consistently identified as the locus of a spatial outlier, suggesting some patterning. A finer spatial scale might reveal more complex spatial patterns. The current data do not allow this to be investigated.