Higgins, Dana: Disaggregating Battle Death Counts using Multiple Imputation

(PDF) The rising popularity and availability of big data has created a push to advance statistical methods that handle very detailed and large quantities of data.  

Within conflict studies, however, the larger problem is not a lack of analytical tools but instead a lack of accurate and detailed data. Gathering data about areas in violent conflict is certainly difficult and necessarily entails a high level of uncertainty about the measures taken.  Rather than ignoring these important covariates or using a theoretically and statistically unjustified estimate, this paper views the disaggregated data as a missing data problem which can be imputed from available data.  This project uses multiple imputation to estimate a count of monthly battle deaths in 141 violent conflicts from 1989 - 2012 using media reports of killings, widely accepted counts of annual battle deaths, and monthly count data released for a selection of conflicts.  Using only this limited information and with 90% confidence, the model can correctly impute monthly counts for 95% of the observations when using the model to impute known information.  Unlike existing approaches or even data collection efforts, these findings additionally have the benefit of providing confidence intervals around the monthly estimates which may be used for inference.