Boundary-Line is the OrdnanceSurvey's vector map of UK administrative and electoral regions. It has all the data needed to locate points in any electoral regions, including CountyElectoralDivisions. The data are supplied in a big FORTRAN-style text file described by a
user manual (which contains a couple of errors
). I have written (perl) code to parse this stuff.
In terms of using this to implement MaPit, we have a choice of running a big list of postcodes through the point-in-polygon tests to annotate an existing postcode database (like the AllFieldsPostcodeDirectory) with the extra fields we need (CED, ScottishConstituency, etc.), or doing the point-in-polygon lookups for each entered postcode and using any postcode-coordinate database. Either is tractable computationally; the latter is, perhaps, a bit more elegant.
One big problem with using BoundaryLine is that we need rather high resolution coordinates. E.g., if we get the accurate-to-100m postcode database, then any house within 50m of a boundary could be misplaced.
Considering wards for a moment, about 80% of the population lives in "urban" areas (see
this ONS discussion of the "Rural and Urban Area Classification 2004"); roughly that number of wards will be urban. Suppose that typical urban wards are circles radius 1km with uniformly distributed population; a 50m band around the edge of the wards will have an area 10% that of the ward itself. Therefore on the 100m dataset we would expect about 8% of the population to be mislocated. With the 10m data, about 1% of the population would be mislocated.
A 10% error rate is not acceptable. A 1% error rate might be.
- CL
So, we now have BoundaryLine from WestSussex in ESRI shapefile format, which is nice (though means I wasted my time writing an NTF parser...). We also have CodePoint (postcodes with 1m coordinates), which should be all the data we need. I'm currently working on the point-in-area tests for this.
