The Jean Goulding Institute funded a novel geobootstrap method for areal interpolation. Areal interpolation is a technique that transfers data values from one set of spatial partitions (sources) to another (targets), which is useful for filling in data gaps. There are various interpolation methods in the literature, and each impose various restrictions to derive unique and stable estimates. These are grounded based on solid theoretical foundations and empirical support and each of these methods has its own strengths & shortcomings. A common distinction is made between those that use ancillary information to spatially constrain the transfer of values within targets (intelligent methods) and those that don’t (simple). However, the general applicability of areal interpolation methods is not well understood because much of the research has focused on interpolating total population counts. Little research has focused on more interesting and perhaps more complicated variables that many applied analysis is likely to require.

This research developed a “simple” geobootstrap method for interpolating income values from the British Population Survey dataset for several local authorities in England. The geobootstrap method extends the nonparametric bootstrap method, which is a resampling technique used to estimate statistics on a population by sampling a dataset with replacement. Instead of assigning an equal probability distribution, where that all sources can be sampled, a distance-based kernel is used to weight the samples, so neighbouring sources have a higher probability of being pooled. In other words, this method takes advantage of the spatial structure of sources/targets and borrows strength from neighbouring units to make estimates.

A key advantage of the geobootstrap method is that interpolations can be made quickly because the geobootstrap does not require any complex data processing to incorporate other ancillary information and can also handle missing data values. This is important when fast decisions are necessary, including responding to emergency disasters. Early results suggest that the accuracy of the geobootstrap method is higher than other simple methods and comparable to intelligent methods.

For a demonstration of the geobootstrap an interactive map is available here