underaggregation

Underaggregation refers to the practice of summarizing or releasing data at a level of detail that is too coarse to capture important variation within the data. In statistical and data-analytic contexts, underaggregation occurs when information is collapsed into broad categories, regions, or units, leading to a loss of granularity and potentially biased or incomplete inferences.

Causes of underaggregation include privacy protections that restrict detail, reporting conventions that limit granularity, computational or storage constraints, and deliberate design choices in data releases. It can occur in various domains, such as social science surveys that report demographics only in large age bands, geographic data summarized at national or regional levels instead of finer local areas, or time-series data that are aggregated over long intervals rather than preserved at higher frequency.

The consequences of underaggregation include missed heterogeneity, attenuated effect sizes, and incorrect conclusions about subgroups or interactions. It can obscure trends, mask differences between subpopulations, and contribute to ecological or aggregation biases when inferences about individuals or smaller units are drawn from overly aggregated data. A related concern is the modifiable areal unit problem, which notes how results can change with different levels of spatial aggregation.

Mitigation strategies focus on increasing data granularity when possible and appropriate, or using model-based approaches that account for within-group variation. Techniques include hierarchical or multilevel modeling, small-area estimation, multiple imputation, data fusion to combine sources with different granularities, and sensitivity analyses to assess how results might change under finer or alternative aggregations.