Written and maintained by CASRAI Editorial Board
Last updated
Short answer: Spatial data analysis is the set of statistical methods for data tied to locations, where nearby observations are usually more alike than distant ones. Which method applies depends on how the data were collected: event locations (point patterns), sampled values at locations (geostatistics) or values for fixed areas (areal or lattice data).
What is spatial data analysis?
The GIS&T Body of Knowledge from UCGIS describes spatial statistics as the field dedicated to describing and modeling georeferenced data. Spatial data analysis, as used in research, is the practice of applying those methods: summarising where things are, testing whether their arrangement is random, and modelling or predicting values across space. The key difference from conventional statistics is that location carries information, so observations cannot be treated as independent of one another.
Three classes of spatial data
The SAS/STAT documentation on spatial point pattern and geostatistical methods separates spatial data into three classes, and this is the most practical way to begin choosing a method:
- Point-referenced (geostatistical) data: values sampled at locations, with the aim of predicting values at unsampled locations.
- Areal (lattice) data: values for a fixed number of areal units, for example crime rates aggregated over counties.
- Point pattern data: a collection of locations of single events of a spatial process, where the interest is in how those events are distributed.
Spatial autocorrelation: why ordinary statistics break down
The UCGIS Body of Knowledge defines spatial autocorrelation as the relationship between values arranged in a spatial structure. Nearby observations tend to be more similar than distant ones, which violates the independence assumption behind conventional statistical tests. Ignoring it can make a model look more certain than the data justify.
The most widely taught measure is Moran’s I. The Bioconductor Orchestrating Spatial Transcriptomics Analysis book describes it as the Pearson correlation between the value at a location and the average of the values at its neighbours. The global value is a weighted average of local values, and local indicators of spatial association (LISA) show each location’s contribution.
That same source makes a point worth carrying into any methods section: the neighbourhood defines the spatial dependency between locations, so analysts should justify their choice of weight matrix. Report how neighbours were defined, because a different definition can change the result.
Choosing a method: a sourced decision table
Every row below comes from the three sources listed at the end of this page (read 2026-11-07).
| Your data / question | Data class | Methods named in the sources |
|---|---|---|
| Are event locations clustered, random or regular? | Point pattern | Quadrat analysis; distance-based tests against complete spatial randomness; Ripley’s K (average number of points within an r-neighbourhood) |
| What is the value at an unsampled location? | Point-referenced (geostatistical) | Kriging, which predicts continuous surfaces |
| Are values for neighbouring areas related? | Areal (lattice) | Global Moran’s I, with LISA for local contributions |
| Does a relationship between variables change across space? | Areal (lattice) | Geographically weighted regression; simultaneous autoregressive (SAR) models |
| Where is the centre and spread of a set of points? | Point data | Centrographic measures: mean center and standard distance deviation |
The SAS/STAT documentation adds that point pattern and geostatistical analyses share corresponding measures to describe correlation among the data. Bayesian spatial models, estimated with MCMC, are also listed in the UCGIS entry for cases where a hierarchical model is needed.
GIS versus spatial analysis
A geographic information system stores, manages and maps spatial data; spatial statistics models data whose values depend on location. In practice the two overlap, because GIS packages include analysis tools, but a map showing a pattern is not a test of that pattern. For GIS tooling see our guides on how to use QGIS and what ArcGIS is used for, and for preparing scanned maps, georeferencing an image in QGIS. For the wider discipline, see what geography is and what remote sensing is.
Is spatial analysis qualitative or quantitative?
The methods in the table are quantitative: they produce statistics, test results and predicted values. Spatial work can also be descriptive, such as visual map interpretation, but the statistical approaches described here are quantitative (an editorial reading of the cited methods).
Software
The Bioconductor book names the R package spatstat as central for point pattern analysis, and spdep for neighbourhood-based measures such as Moran’s I. For an introduction to the language, see what R is used for in research. Whichever package you use, record the version and the weight-matrix specification so the analysis can be reproduced.
Common misunderstandings
- A map is not an analysis. Visible clustering on a map needs a formal test against randomness.
- Autocorrelation is not just a nuisance. It is also a finding: it describes how strongly values depend on location.
- The weight matrix is a modelling choice. It should be justified and reported, not left at a software default without comment.
- Standard errors from ordinary methods can mislead when residuals are spatially dependent, because the independence assumption does not hold.
Frequently asked questions
What is spatial data science?
Usage varies and there is no single standard definition. This guide treats it as spatial analysis plus the programming and data-handling workflow around it, which is an editorial reading rather than a formal definition.
What is the difference between GIS and spatial analysis?
GIS is the system for storing, managing and mapping spatial data. Spatial analysis, in the statistical sense, models and tests data whose values depend on location. Many GIS packages include analysis tools, so the two are often used together.
What are the 5 components of spatial analysis?
There is no single standard list of five. The sources used here organise the field by three data classes (point pattern, geostatistical, areal) and the method families that fit each, as in the table above.
What is spatial statistics?
Spatial statistics is the field dedicated to describing and modeling georeferenced data (UCGIS GIS&T Body of Knowledge). Its methods include centrographic measures, point pattern analysis, geostatistics, areal methods and Bayesian spatial models.
Is spatial analysis qualitative or quantitative?
The statistical methods covered here are quantitative. See the section above.
Sources
- SAS Institute, SAS/STAT documentation, spatial point pattern and geostatistical methods overview: support.sas.com (read 2026-11-07).
- UCGIS, GIS&T Body of Knowledge, Spatial Statistics (AM-03-011): gistbok-ltb.ucgis.org (read 2026-11-07).
- Bioconductor, Orchestrating Spatial Transcriptomics Analysis, Spatial statistics: bioconductor.org (read 2026-11-07).
