Global Impact: Statistical Analyses of Conflict Data in Syria, Guatemala, and Colombia

• Activity Number: 374
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Committee on Scientific Freedom and Human Rights
• Author(s): Shira Mitchell*+ and Al Ozonoff and Kristian Lum and Alan M. Zaslavsky and Brent Coull
• Companies: Harvard and Harvard and Virginia Tech and Harvard and Harvard School of Public Health
• Keywords: Population size estimation ; capture-recapture ; multiple record-systems estimation ; missing data ; structural zeros ; hierarchical models

Abstract

Since 1964, tens of thousands of people have died in Colombia's armed conflict. Underreporting the level of violence obscures the true nature of the conflict, precluding development of effective solutions. We develop hierarchical log-linear capture-recapture models to estimate the number of armed conflict killings that occurred in Casanare, Colombia in the years 1998-2007. Lack of data the early years motivates the use of hierarchical models that borrow strength across time. We investigate two methods to handle groups actively collecting data in different but overlapping time-periods. One fills in the inactive periods, treating counts in those years as missing data. Another does not, instead incorporating the inactivity into the model. We compare these, as well as hierarchical versus unpooled models. A simulation study shows that the Bayesian hierarchical models have shorter confidence interval width, with similar or better coverage than the unpooled models. They give useful intervals for the number of killings in the early years, where there are less data, so we can look at trends across time that guide political analysis of the conflict.