A Fast Spatial Multiple Imputation Procedure for Imprecise Armed Conflict Events

Abstract:

The proliferation of large, geographically disaggregated data on armed conflict, protest and other forms of political violence has led to a substantial development of research avenues. It has also brought with it numerous problems, both in terms of management of data quality and assurance of consistency. Dataset authors are generally aware of shortcomings in the sources they use, and alert users to such limitations. Still, researchers have been slow to provide solutions to these problems, even where they are clearly stated by dataset authors, in many cases due to the computational costliness of the canonical solutions. One such major but mostly ignored problem is the presence of large amounts of “known geographically imprecise” (KGI) or “known temporally imprecise” (KTI) observations in important datasets. In the spatial domain (KGI), dataset authors typically alert to geographical imprecision for 20–40% of observations . We introduce a simple, multiple-imputation based technique to address this issue. We show that the resulting imputations are reasonabley precise, and that using the imputed data considerably improves out-of-sample predictive performance relative to excluding the imprecise observations.

Authors:

Mihai Croicu and Håvard Hegre