Spatio-Temporal Models for Point Pattern Data with Network-Dependent Sampling

• Activity Number: 173
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Tyler H. McCormick*+
• Companies: University of Washington, Seattle
• Keywords: Bayesian analysis ; Non-homogenous Poisson Process

Abstract

According to a 2007 special issue of The Lancet, less than one-third of the world's population is covered by accurate information on births and deaths. In such areas, data about vital indicators are collected using surveys that gather information about births and deaths indirectly via reports from community and family members. In this talk, I demonstrate that this data collection technique is a form of preferential sampling (Diggle 2003) where the sampling mechanism depends on an individuals' social network. I introduce a Bayesian non-homogenous Poisson process model which accounts for the both the network-dependent sampling mechanism and non-sampling errors introduced as respondents attempt to recall events far in the past. I demonstrate this model using data from the Demographic and Health Surveys (DHS) and explore extensions of this modeling framework which incorporate spatial dependence to produce subnational and regional estimates of fertility and mortality.