Disparitiescan

Disparitiescan is a concept and software tool designed to systematically audit datasets and decision-making pipelines for discriminatory disparities. It focuses on identifying, quantifying, and reporting differences in outcomes across demographic groups defined by protected attributes such as race, gender, age, disability, or ethnicity. The goal is to support fairness assessments in machine learning, analytics, and policy evaluation.

Disparitiescan ingests data that includes outcomes and sensitive attributes, then computes a range of fairness metrics and conducts subgroup analyses. Typical metrics include demographic parity difference, disparate impact ratio, equal opportunity difference, and predictive parity, among others. The tool can evaluate model performance across different decision thresholds, calibration, and error rates for each group, and it often produces dashboards and drill-down reports to help stakeholders interpret results.

Applications of disparitiescan span both industry and public sectors. It is used to audit hiring and promotion pipelines, lending and credit scoring, healthcare analytics, and criminal justice workflows. Organizations employ it to support compliance with anti-discrimination laws, inform remediation strategies, and guide data collection, feature engineering, or model adjustments aimed at reducing unfair disparities.

Limitations and caveats are important considerations. The quality and availability of protected attributes strongly affect results; missing data and sampling biases can distort findings. The presence of unfairness metrics does not prove causality, and improvements in one fairness dimension can affect others or reduce overall accuracy. Privacy concerns and legal constraints also shape how disparitiescan analyses are conducted and shared.

Related concepts include algorithmic fairness, disparate impact, bias auditing, and various fairness metrics used in data science.