Modeling Sparse Demographic Data with Random Effects: Studies in Migration Modeling and Small Area Estimation

Nicholas Nagle, University of Tennessee

The increasing dimensionality of data is introducing challenges to how we analyze and visualize data. This is compounded by the relative sparsity of these new data. For instance, survey data for age-specific migration flows between places contains many more zero-counts than counts of positive flow. Random effects have been shown to be a useful technique for modeling sparse data because they allow the opportunity to "borrow" information across observations. Two case studies are demonstrated showing the potential of random effects models in applied demography. The first develops an age-specific model of migration into and out of PUMAs in Tennessee. The second develops a model for creating small area estimates from public use microdata.