Urban planning, estate management, architecture
Planning and the built environment
Location is a variable. ZStat treats it as one, rather than as a column of numbers that happen to be coordinates.
A regression on property prices with no spatial term treats two flats on the same street as independent observations. They are not, and every standard error it reports is too small as a result. These procedures are the ones planning, valuation and architectural research actually use.
What you can do
Moran's I gives one answer for the whole study area: clustered, dispersed, or indistinguishable from random. Tested by permutation rather than an approximation that assumes what is under test.
The Getis-Ord Gi* statistic answers the same question at every location, which is what produces a hot-spot map. A study can show no global pattern and still contain one dense cluster that matters enormously.
The nearest neighbour index for boreholes, kiosks, clinics, filling stations - where there is no value attached, only the positions.
Hedonic price modelling through regression: which attributes of a property, and by how much, with the assumption checks that decide whether the model holds.
Questionnaire design, reliability, factor analysis and group comparison for satisfaction, walkability and accessibility work.
What it refuses
- Treating latitude and longitude as plain x and y. A degree of longitude is 111km at the equator and 96km at Kano; distances are computed on the sphere.
- Hiding that a hot-spot map runs one test per location, so on a hundred places about five look significant with no pattern at all. What matters is whether they are next to each other.
- Pretending it can tell whether your coordinate columns were swapped. In Nigeria both fall in the same numeric range, so nothing detects it and the result says so.