Presentation: Combining Data on Different Spatial and Temporal Scales
Speaker: Katherine Wilson, Graduate Student, UW Biostatistics
Abstract: Many developing countries lack vital registration systems that provide detailed information on health outcomes. Instead, data used to estimate risk over space and time can come from complex household surveys and population censuses. These sources often do not provide the same level of information, prompting the need for new statistical methods. In this talk, I focus on two such scenarios. In the first, the interest is on incorporating census data that has been aggregated over regions into a model where health risk varies continuously in space. In the second, the interest is on combining two types of birth history data to model the under-five mortality rate. One source contains detailed information on when births and deaths of children occurred by mother, whereas the second contains only the total number of births and deaths by mother. Both of these examples deal with the question of how to construct models when the underlying process is at a different resolution than the reported data. In the spatial literature, this is known as the change of support problem.