Confidence intervals in small cluster surveys of conflict mortality: how sure are we really?

Michael Spagat, University of London
Gabriela Guerrero-Serdan, University of London

We analyze the methods used for the calculation of confidence intervals (CI’s) in cluster surveys of conflict mortality, proceeding along both theoretical and empirical tracks. For the theory we apply the methods developed in Rosenblum and van der Lann (2008). This paper argues that CI’s based on the central limit theorem or bootstrapping can be highly misleading in small samples. They propose alternative methods which, when applied to the second Lancet survey of Iraq mortality, produces considerably wider CI’s than the original ones. We will recalculate CI’s for some other small-sample conflict mortality surveys that have appeared in the literature using Rosenblum-van der Laan methods. Then we will use simulation techniques based on cluster-level data drawn form large conflict mortality surveys, including the ILCS and the IFHS which were both done in Iraq, to further explore the sampling properties of small cluster surveys.