Disaggregation of population forecasts for school children in Berlin
Lorena Gril1, Prof. Ulrich Rendtel1, Felix Skarke* 1
Abstract
In Berlin, neighbourhoods (called ”lifeworld-oriented spaces” LOR) are the spatial basis for planning, forecasting and monitoring demographic and social developments. These geographical units are nested in larger area systems called planning spaces (PLR), district regions (BZR) and forecasting spaces (PGR). On the PGR level the Berlin senate forecasts the city’s population, but for planning purposes forecasts for school children aged 6 to 11 need to be disaggregated to school planning regions (SPR), which are not hierarchically nested in the forecasting level areas. One way to produce the desired disaggregated estimates at SPR level is the use of Kernel Heaping algorithm. The method, initially designed for iterative kernel density estimation with aggregated data, also solves the so-called support problem by generating pseudo-samples to make switching between non-hierarchal area systems possible.
The question arising in this context is whether the results of the Kernel Heaping method can be further improved. Therefore an approach is proposed, that makes use of Small Area Estimation (SAE) techniques to refine the estimates of Kernel Heaping. In SAE area-level models, information is used at a specific area level. However, if the area levels of the dependent and independent variables are different, it is not obvious how regression modeling should be done.
Our proposed method combines two pertinent methods of official statistics, the Fay-Herriot Model and the Kernel Heaping algorithm. The Fay-Herriot model employs a combination of direct and regression estimation methods to achieve precise results. Traditionally the direct estimators used in the model are design-based estimators, that only use the information in the area of interest. The new method switches from established direct estimators to the Kernel Heaping algorithm, because it has the advantage in situations where the dependent variable and auxiliary information are not measured at the same geographical level. The approach is applied to real world data from Berlin.