Estimating the intensity of conflict in Iraq

Over the last five years several websites have appeared that track the number of deaths of coalition personnel in the current Iraq conflict. This article proposes using a recently developed multiscale variance stabilization method (the datadriven Haar-Fisz transform) to obtain good estimates of the mean intensity of deaths and demonstrates its statistical advantages over both the running mean (currently used on some websites) and also the Box-Cox transformation. The article analyses both the number of deaths due to all causes and also due to hostile actions between June 2003 and March 2006. In both cases it appears that the marginal variance of both time series is an increasing function of the marginal mean. Further, although our estimates do exhibit peaks related to local increases in conflict intensity there appears to be clear underlying increase in conflict intensity during the conflict but with a decline and levelling-off since June 2005. The article also proposes a tentative inverse relationship between deaths due to non-hostile actions and deaths due to hostile actions.