Requirements Analysis

This page illustrates some classification schemes for Spatial Statistical tests to be included in the GeoMed software. The details of the implementation and uses of the various tests are pulled from Leah Estberg's Requirements Analysis document (provided in Word or html format).

See also Leah's data file format specification document for individual tests.

We're working on a statistical advisor.

Cluster Detection Method Hierarchical Classification Scheme

This just shows one "taxonomy" of the tests: another is based on Information Frames.

Global: Cuzick & Edwards Grimson's Method Moran's I Moran's I adjusted^* Ripley's K function Besag & Newell's Method Rogerson's method	Space
Local: Local Moran (LISA) Turnbull's Test Kulldorff's Spatial Scan Statistic
Focused: Bithell's linear risk score method Diggle's method Score test of Lawson and Waller Stone's Test^**

Space-Time

Grimson's method
k-Nearest Neighbor
k-Nearest Neighbor adjusted^*
Knox's method
Knox's method adjusted^*
Kulldorff's Spatial Scan Statistic
Mantel's method
Mantel's method adjusted^*

Time

Dat's method
Ederer-Myers-Mantel method
Empty Cells
Grimson's method
Larsen's method
Scan Test
CuSum

These cluster detection methods are used to investigate disease clusters in space (cases of disease tend to aggregate over particular locations) or in time (cases of disease tend to aggregate over particular time periods). Some methods can detect disease clusters in space that change over time, when there is space-time interaction (Space x Time).

Some specialized types of temporal clustering methods are called surveillance methods. These methods are used to continuously monitor disease frequency data as they are updated with recent disease events.

Spatial cluster detection methods can be classified in several different ways. "Global" tests detect clustering throughout the study region without interest in the specific locations of the clusters. "Local" and "focused" tests are similar because they are used to detect clustering in localized areas. Specifically, a "focused" test is used detect clustering around a point source exposure to a factor that is proposed to increase risk of disease. For that reason, "global" and "local" methods can also be considered "general", in the sense that they look for clustering without pre-determined hypotheses about location of a cluster. It may be useful to consider a second, tabular classification scheme for the spatial cluster detection methods.

*Some of the methods that don't account for the size of the population at risk have adjusted versions for incorporating information about population sizes that may vary by location or change through time.

**We have chosen not to implement this method: according to Dr. Geoff Jacquez, "We elected to include the Score test that is attributed to Lawson and Waller since it has been demonstrated in both theoretical and applied settings to have better statistical power than Stone's test." In other words, better methods exist.

Website maintained by Andy Long. Comments appreciated.

aelon@sph.umich.edu