Civil wars rarely span throughout the entire territories of host countries. More frequently, they are confined to specific sub-national regions while other parts of the countries are relatively peaceful. In order to avoid the ecological fallacy, or explaining local phenomena with country-level measures, an increasing number of quantitative studies of civil war are applying disaggregated data and research designs. Although such approaches are certainly promising, they introduce or emphasize problems related to classical statistical inference. A fundamental underlying assumption is that of independence, that the units of analysis are unrelated to each other. With geographical data, this assumption is questionable, and particularly so for disaggregated observations. Disaggregation also presents another challenge, known as the modifiable areal unit problem (MAUP). In this paper, we investigate possible effects of spatial autocorrelation and MAUP by conducting a series of disaggregated analyses on the outbreak of civil war in Africa, 1970â€“2004. To test for the zoning effect of MAUP, we use two grids with identical resolution (100 x 100 km) but with different zoning (52% overlap). The scale effect of MAUP and the influence of spatial autocorrelation are explored by estimating the same regression model on four alternative samples, generated from grids with resolutions of 50, 100, 150, and 200 km, respectively.
Rød, Jan Ketil & Halvard Buhaug (2008) Civil Wars: Prospects and Problems with the Use of Local Indicators, presented at 49th Annual Convention of the International Studies Association, 26–29 March.