|
Abstract:
|
Methods have been developed for imputing values that are partially-missing or censored. These are typically found in the survival statistics literature, and often using hazard functions over continuous time intervals with known bounds. Here, we present a method for imputation of discrete values that are estimated to be within a certain interval, but otherwise unknown and missing, in a Bayesian framework. We use the given interval only to create informative priors for each missing value, and also incorporate the nested structure of the different types of casualties within the total number of casualties. We apply this method to the number of casualties caused by US drone strikes in the Middle East. The data set contains the locations and dates of drone strikes since 2001, and includes the total number of casualties caused by each drone strike. The number of those casualties that were civilians or children are also given, as well as the total number of injured persons. Most of these death counts are given only up to a known interval. We compare this method to several other imputation methods, and give extensions to future work on using the complete imputed dataset.
|
