Aggregation granularity must be set by the sparsest region, not average density

What it is

Release granularity for spatial data cannot be set from overall average density. It must be checked against the sparsest areas, times, and allowed filter combinations. A high average per grid cell can conceal edge areas with only one or two records; those minimum cells constrain the privacy threshold.

Why it happens

Density is often highly uneven. High counts in city centres lift the average while hiding empty and tiny cells in rural areas, at night, or for rare categories. Attackers do not target the average place; they seek the easiest cell to link. Fixed grids can also place the same person in several low-count tables under different filters, further narrowing the candidate set.

Where it stops holding

Uniformly coarsening an entire map to protect its sparsest areas can needlessly erase urban analytical detail. Adaptive aggregation—merging neighboring cells or using hierarchical geographic units—can help, but readers must understand that different cells have different comparison limits. Minimum count is not the only risk measure; sensitive-attribute homogeneity and differencing still matter.

Applying it

• Calculate cell-count distributions under every allowed time, category, and region filter. Decide against the minimum, not the mean.
• Recursively merge low-density cells or move them to a higher administrative level, while making changed boundaries and aggregation levels clear.
• Verify the released view with low-count, differencing, and multi-filter tests. Block or aggregate again wherever a route creates a below-threshold cell.

Related

• Same group: U12.06.1 Precise coordinates can reveal homes and daily routines · U12.06.2 Aggregating to areas can reduce re-identification risk · U12.06.3 Sparse regions can remain uniquely identifying after aggregation · U12.06.4 Location jitter introduces false conclusions at small scales
• Nearby: U12.06.3 Small-cell risk follows the smallest candidate set · U8.05.2 Small counts should not be published as precise values
• Search terms: adaptive aggregation · minimum cell count · privacy threshold