Tackling Documentation Debt: A Survey on Algorithmic Fairness Datasets

Abstract

A growing community of researchers has been investigating the equity of algorithms, advancing the understanding of risks and opportunities of automated decision-making for historically disadvantaged populations. Progress in fair Machine Learning (ML) hinges on data, which can be appropriately used only if adequately documented. Unfortunately, the research community, as a whole, suffers from a collective data documentation debt caused by a lack of information on specific resources (opacity) and scatteredness of available information (sparsity). In this work, we survey over two hundred datasets employed in algorithmic fairness research, producing standardized and searchable documentation for each of them. Moreover we rigorously identify the three most popular fairness datasets, namely Adult, COMPAS, and German Credit, for which we compile in-depth documentation. This unifying documentation effort targets documentation sparsity and supports multiple contributions. In the first part of this work, we summarize the merits and limitations of Adult, COMPAS, and German Credit, adding to and unifying recent scholarship, calling into question their suitability as general-purpose fairness benchmarks. To overcome this limitation, we document hundreds of available alternatives, annotating their domain and the algorithmic fairness tasks they support, along with additional properties of interest for fairness practitioners and researchers, including their format, cardinality, and the sensitive attributes they encode. In the second part, we summarize this information, zooming in on the domains and tasks supported by these resources. Overall, we assemble and summarize sparse information on hundreds of datasets into a single resource, which we make available to the community, with the aim of tackling the data documentation debt.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 13 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.