Accuracy of areal interpolation methods for count data

TL;DR

This paper evaluates the statistical accuracy of areal interpolation methods for count data, focusing on proportional weighting and Poisson regression, revealing no universally superior technique.

Contribution

It introduces a stochastic Poisson model to compare areal interpolation methods for count data, highlighting their dependence on data characteristics.

Findings

No single method always outperforms others.

Accuracy depends on variable nature and correlation.

Poisson model provides a framework for evaluation.

Abstract

The combination of several socio-economic data bases originating from different administrative sources collected on several different partitions of a geographic zone of interest into administrative units induces the so called areal interpolation problem. This problem is that of allocating the data from a set of source spatial units to a set of target spatial units. A particular case of that problem is the re-allocation to a single target partition which is a regular grid. At the European level for example, the EU directive 'INSPIRE', or INfrastructure for SPatial InfoRmation, encourages the states to provide socio-economic data on a common grid to facilitate economic studies across states. In the literature, there are three main types of such techniques: proportional weighting schemes, smoothing techniques and regression based interpolation. We propose a stochastic model based on Poisson point patterns to study the statistical accuracy of these techniques for regular grid targets in the case of count data. The error depends on the nature of the target variable and its correlation with the auxiliary variable. For simplicity, we restrict attention to proportional weighting schemes and Poisson regression based methods. Our conclusion is that there is no technique which always dominates.