Meshblock mysteries

By Aaron Schiff 28/01/2015

Warning: This post is unlikely to interest many people.

Meshblocks are the basic unit of geographically reported data in New Zealand. They are the smallest geographic areas at which Stats NZ makes data available outside of the datalab.

There are three main sources of information about meshblocks:

1. Geographic area files – These are tables that show the concordance between meshblocks and larger geographic areas such as area units, territorial authorities, and regions.

2. Geographic boundary files – These provide data for drawing maps of meshblocks (and other geographic areas) in GIS software. There are two basic sets of shape files: “full” shapes that extend into the sea and include things like uninhabited islands, wharves and marinas, and “clipped” shapes that stay within the coastline.

3. Various datasets that report information at the meshblock level, such as the 2013 Census meshblock dataset.

When working with these files, I’d noticed that they contain different numbers of meshblocks, and hence some meshblocks are in some files but not others. (Another complication is that meshblocks change over time as the geographic distribution of the population changes, but that problem can largely be avoided by using data files all from the same year, if possible).

These discrepancies have always bothered me slightly, so I sat down to work out exactly what are the differences in the sets of meshblocks in the various files. Here’s a summary for the 2013 files (BTW I can’t believe it’s 2015 and the quickest way for me to add a tidy table to this post is to take a screenshot of Excel):


So from this you can see the Areas Table is the master list — every meshblock in any of the other files is in the Areas Table. But there are meshblocks in the Areas Table that are not in the other files. In particular there are 8 meshblocks in the Areas Table that are not in any other files. Meshblocks don’t have names or descriptions, so all you can do to identify these 8 is look up their corresponding area unit (which could be a larger area) in the Areas Table (area units have descriptive names). The area units corresponding to these 8 meshblocks are named:

Ross Dependency
NZ Economic Zone
Oceanic-Bounty Islands
Bounty Islands
Oceanic-Snares Islands
Snares Island
Oceanic-Antipodes Islands
Antipodes Islands

So these are some oceanic areas that belong to NZ, some uninhabited islands, and New Zealand’s bit of Antartica. It makes sense that these don’t appear in the Census meshblock dataset, although I’m not sure why they don’t appear in the “full” shape files.

The most significant difference in the meshblock sets, if you are analysing the Census meshblock dataset like I have been, is the meshblocks that are in the Census dataset but not in the shapes files. There are 640 meshblocks missing from the “clipped” shapes and 8 missing from the “full” shapes. It turns out that the latter 8 are a subset of the 640, so these 8 are missing from both types of shape files. Again, no names for these and we can’t map them, so here are the corresponding area unit names:

Oceanic-Kermadec Islands
Kermadec Islands
Oceanic-Oil Rigs Taranaki
Oceanic-Campbell Island
Campbell Island
Oceanic-Oil Rig Southland
Oceanic-Auckland Islands
Auckland Islands

I’m not sure why the “oceanic” areas in this list deserve to be in the Census dataset but some in the list above did not. Let’s just it put it down to life not being fair. Anyway, the others make sense: oil rigs are inhabited (I suppose), and the other islands either have people there sometimes or have been inhabited in the past.

The remainder are the 632 meshblocks that are in the Census dataset and in the “full” shape files but not in the “clipped” shapes. Too many to list, but they can be mapped. The grey areas are the meshblocks in both shape files and the red areas are only in the “full” shapes:


The last question is whether these 632 meshblocks are very important for Census analysis. The 2013 Census reports 510 occupied private dwellings in these areas (out of 1.56 million total) and a usually resident population of 753 (out of 4.24 million total). So no, not very important and we can safely make maps with just the “clipped” shapes.

Phew, I feel better now.