Comparing three statistical techniques for Space-time clusters with county-level fertility data from Costa Rica.

Population projections at a small area level are sensitive to random departures in temporal trends of population change components. In order to have robust trends for the county level projections in Costa Rica, the team in charge of forecasting decided to determine county clusters based on the past trends. In this paper, we compare three descriptive techniques used to build space-time clusters for the Crude Birth Rate: Functional Data Analysis (FDA), the scan statistic, and Geographically Weighted Regression GWR. GWR was the technique that yielded clusters related to the diffusion paradigm for fertility decline. However, the scan statistic provided a more parsimonious set of clusters that are more tractable for population projections.