Stochastic model-based methods for handling uncertainty in areal interpolation

Abstract
The trend towards more fully integrated Geographic Information Systems remains hampered by the recording of different groups of variables with respect to mutually incompatible sets of areas. We describe a method developed for interpolation of agricultural census employment counts from a source set of areas (parishes) to a target set (pseudo postcode sectors). This was done by introducing a third set of statistical reporting areas (namely, output areas for the British population census), utilised because it is approximately nested within both source and target sets. Interpolation of the count data from parishes to output areas was stochastic, following a model for the probability that employed individuals were employed in agriculture, but obeying the constraint that the total number of agricultural employees per parish should remain constant. The interpolation procedure was implemented within a Bayesian statistical framework using Markov chain Monte Carlo methods, enabling all sources of uncertainty in the model to be taken into account. Each interpolation was summed to create a new set of counts aggregated to the pseudo-postcode sector level. The simulated values at this level are presented both via summary statistics and as individual realisations to reflect the uncertainty in interpolation.
Year
2013
Category
Refereed journal
Output Tags
SG 2006-2011 P3 Environment - Miscellaneous