Choosing a problem and collecting the data
What makes a problem workable
- It must be real, at a given location. The standard rules out simulations. A real problem someone would actually want solved is the safest choice.
- It must have a spatial dimension. The answer must depend on where things are. If moving everything 500 m to the left would not change the answer, it is not a spatial problem.
- It must be answerable with the data you can get. Check the layers exist before you commit. A brilliant question with no available data becomes three weeks of frustration.
- It must be small enough to finish. One town, one question, one threshold. Improving public transport in the region is a project; which dwellings in this town are beyond a 400 m walk of a stop is an assignment.
Problem types that work
- Accessibility. Who can reach a service — a bus stop, a school, a park, a defibrillator, a supermarket — within a set distance or time.
- Suitability. Where the best site is for something, given a set of requirements. Where should a new playground go, given that it must be more than 100 m from an arterial road and within 400 m of at least 150 dwellings?
- Exposure. Which features fall inside a mapped zone — a flood overlay, a coastal hazard line, a noise contour.
- Change over time. What has appeared, disappeared or moved between two dated layers of the same place.
- All four produce a countable answer, which is what makes them defensible. A problem whose answer is an adjective is the wrong problem.
Collecting spatial data: the two routes
- Route 1 — fieldwork with a spatial component. You collect it yourself, and every observation carries a location: a GPS point, a grid reference, a marked position on a printed map.
- The rule that makes fieldwork usable: record the location at the same moment as the observation. A count with no position cannot enter a GIS.
- Good fieldwork layers: counts at recorded points, condition scores along a transect, photographs with coordinates, timed walking routes.
- Route 2 — accessing spatial data from other sources. The standard allows this outright, and for most problems it is the practical route.
- Council open data portals — roads, reserves, zoning, rating units, stormwater.
- LINZ Data Service — topographic layers, property boundaries, addresses, aerial imagery.
- Stats NZ — statistical area boundaries and the figures attached to them.
- Regional council layers — hazard overlays, monitoring sites, transport networks.
- Mixing the two is usually strongest. Secondary layers give you the town; your fieldwork gives you the variable nobody else has measured.
Recording the data properly
- Every layer needs three things recorded before you use it: what it is, where it came from, and when it was current. Put this on the layout at the end, and keep it in a notes file from the start.
- A layer's date matters more than students expect. A stop network from 2019 answers a 2019 question. If your problem is about now, say when your data was current and what might have changed.
- Check the layer against reality on one or two features. Does the bus stop the layer puts on the corner actually exist? One field check catches an entire category of error and takes twenty minutes.
- Note what is missing. If the dwelling layer omits a new subdivision, say so, and say which direction that pushes your result.
Sufficiency: the Merit word
- Merit asks for sufficient data. Sufficient means enough to support the conclusion you are drawing, and it is judged against your question, not against a fixed number.
- For a proportion, the layer must be complete rather than sampled. 69.3 per cent of dwellings is only meaningful if the dwelling layer holds all of them.
- For a comparison, the two things being compared must be measured the same way. Comparing a hand-counted 2026 field survey against a 2019 automated layer measures your method as much as the town.
- More data is not automatically better. A precise layer of the wrong thing is not sufficient; a modest layer of exactly the right thing is.
Worked Example
Worked example
Set up the invented Puketawa problem completely, from question to data inventory, so that the analysis could begin.
Answer:
Step 1 — state the problem as a question with a countable answer.
How many of Puketawa's dwellings are more than a 400 m walk from an existing bus stop, and where should two new stops go to reduce that number the most?
Two countable outputs: a number of dwellings, and two locations. Neither can be answered by looking.
Step 2 — check it against the three tests.
Real and located? Yes — one named town and its actual stop network. (Puketawa is invented for this page; the real version of this task uses a real town.) Spatial? Yes — the answer is entirely about relative position. Small enough? Yes — one town, one threshold, one service.
Step 3 — state the parameters, and justify each.
- Threshold: 400 m. A commonly used planning figure for a comfortable walk to a stop. Stated up front, because it changes the answer.
- Distance measured along the street network, not straight-line, because people walk on streets.
- Unit counted: dwellings, not people, because a dwelling layer is available and a resident-level layer is not.
Step 4 — build the data inventory.
| Layer | What it holds | Source type | Currency | Coordinate system |
|---|---|---|---|---|
| Aerial imagery | Basemap for context and checking | Council imagery service | 2025 flight | NZTM 2000 |
| Road centrelines | Streets, for network distance | Council road layer | Updated 2026 | NZTM 2000 |
| Dwelling points | 1,740 points, one per dwelling | Council rating units | Extracted June 2026 | NZTM 2000 |
| Bus stops | 9 points | Regional council transport data | Extracted June 2026 | NZTM 2000 |
Step 5 — check the layers before trusting them.
Walk to two of the nine stops and confirm they exist where the layer puts them. Check the dwelling layer against one street counted on foot: the layer shows 34 dwellings in that street, and the field count finds 35 — one new build not yet in the rating data. Note it: the layer slightly undercounts recent construction, which will make my coverage figure marginally optimistic for the newest part of town.
Step 6 — confirm sufficiency.
The dwelling layer is complete rather than sampled, so a proportion is meaningful. Nine stops and 1,740 dwellings give a stable result. All four layers share NZTM 2000, so they will align without reprojection.