Space-Time Local Indicators of Spatial Association (ST-LISA)

Standard LISA is computed on a single cross-section of spatial data and identifies clusters or outliers at one point in time. ST-LISA applies the same local statistics across multiple time periods and then characterizes each location by its temporal pattern of clustering — stable, emerging, fading, or sporadic — revealing dynamics that a single snapshot cannot capture.

Because you compute a separate significance test for every location-time combination, the chance of false positives accumulates rapidly. Common corrections include Bonferroni (conservative), Benjamini-Hochberg false discovery rate (less conservative), or setting a stricter alpha (e.g., 0.01 instead of 0.05). Report the correction method used and interpret borderline results cautiously.

Yes, by comparing the cluster maps across consecutive time periods you can identify units that transition from non-significant to high-high (emerging cluster edge) or from high-high to non-significant (retreating cluster). Some implementations formally classify each unit by its transition pattern across the full observation window.

Queen contiguity (shared boundary or vertex) is standard for areal data such as administrative regions. For point data, distance-band or k-nearest-neighbor weights are common. Use a row-standardized matrix so that location-level statistics are comparable across units with different numbers of neighbors. Sensitivity-check your results under two or three different weight definitions.

There is no strict minimum, but fewer than 3–4 time periods typically provides insufficient evidence to characterize temporal patterns. In practice, studies in epidemiology and criminology commonly use 10–50 time periods (e.g., years or months) to distinguish stable from episodic clustering.