Underdata

Underdata is a term used to describe datasets that are incomplete, underrepresented, or insufficient to support reliable conclusions. It denotes a condition where data coverage, granularity, or quality falls short of what is required for robust analysis, often due to gaps, biases, or privacy constraints. Although not a formal statistical term, underdata is used in data governance, journalism, and risk assessment to highlight limitations in evidence.

Characteristics of underdata include missing values, sparse sampling, biased coverage across time, space, or populations, small sample sizes relative to the research question, and high uncertainty in inferences. Underdata can arise when data collection methods exclude certain groups, when privacy protections limit data detail, or when data silos prevent integration of complementary sources.

Causes commonly involve budget constraints, regulatory or ethical restrictions, technical failures, nonresponse or attrition, and deliberate or incidental data curation choices that favor certain segments over others. The impact of underdata can be substantial, leading to biased estimates, overgeneralization, misinterpretation of trends, and poor decision-making if acknowledged limitations are not communicated.

Mitigation strategies focus on reducing gaps and quantifying uncertainty. Approaches include data augmentation or imputation where appropriate, leveraging related data sources to borrow strength, using statistical models that account for missing data, conducting sensitivity analyses, and clearly reporting data limitations. Targeted data collection efforts and transparent documentation are also important to improve the evidentiary value of analyses affected by underdata.

Examples of underdata appear in public health with rare diseases, climate research with sparse sensor networks, and market research that underrepresents certain demographics. See also: missing data, data quality, data sparsity, data imputation, data coverage.