UK Postcode Polygon Accuracy Comparison Part 2

So we have seen from the previous comparing the raw polygon accuracy between Voronoi generated polygons and NRS generated postcode polygons: Results.

The physical results are interesting, and a visual examination can provide a useful overall comparison, but how does this actually impact me?

I have a CAG from GCC and I just want to attach a postcode to it. How different will my results be between a true postcode boundary dataset from the NRS, and a generated Voronoi dataset from the OS?

I’m glad you are still with me, it might be useful to explain how postcodes actually work in this context:

Lets take a postcode of G31 2XT how does it break down?
Area: G
District: G31
Sector: G31 2
Unit: G31 2XT

So then we can compare how an actual address dataset, like the Glasgow CAG, spatially joined to two postcode datasets compare:

Assuming the NRS dataset is correct (a good assumption) how accurate is a postcode based on an OS Code-Point Open generated Voronoi polygon based on Glasgow City Council residentially classified properties as of 16/11/2014:

Total number of properties:
245096     100%
Correct Area:
245096     100%
Correct District
243650     99.4%
Correct Sector
240956     98.3%
Correct Unit
174344     71.1%

We can see that up to a sector level a Voronoi polygon can produce an extremely accurate results. A visual comparison of how this plays out in Glasgow can be seen here, with the legend best read from the bottom:

UK Postcode Comparison

3 thoughts on “UK Postcode Polygon Accuracy Comparison Part 2

  1. JasonDW

    Great analysis. Super interesting. But how does the OS Codepoint Open team determine the location of their postcode unit points? Each of those initial point placements is critical in Voronoi polygon construction.

    1. Heikki Vesanto Post author

      I believe they are the polygon centroids from the OS Codepoint Polygon dataset, which contain “actual” postcode areas.

  2. Alasdair


    I’m planning on generating postcode sector and district boundaries from Code-point Open. Have you written more about your approach to this process? I’ve been experimenting generating voronoi boundaries from the individual postcode unit locations, however they are rounded edged and am not sure how well they will translate to the often complex district and sector boundaries.

    Any thoughts or suggestions would be appreciated, we are unable to afford the licensing costs of Code-point with Polygons.


Leave a Reply

Your email address will not be published. Required fields are marked *