Source-linked AI summary
Towards a Critical Race Methodology in Algorithmic Fairness
Alex Hanna, Emily Denton, Andrew Smart, Jamila Smith-Loud
TL;DR
Algorithmic fairness research often treats race as a fixed attribute, despite racial categories being socially constructed and tied to structural inequality. The paper synthesizes critical race theory, racial-classification history, and lessons from biomedical, public-health, and survey research to develop a more contextual methodology. It concludes that fairness research should recognize race’s multidimensionality, scrutinize its conceptualization and measurement, focus on racism’s social processes, and include affected communities’ perspectives.
Problem
Fairness frameworks often encode race as a fixed attribute even though racial categories are contextual, unstable, and socially constructed, potentially minimizing structural aspects of algorithmic unfairness.
Method
The paper combines critical race theory and sociological work with histories of racial classification and lessons from biomedical, public-health, and survey research.
Results
The paper argues that fairness research should account for race’s multidimensionality, take conceptualization and operationalization seriously, focus on processes producing racial inequality, and consider affected communities’ perspectives.
Takeaways & Limitations
Race-based fairness analysis should be grounded in the social processes and contexts through which racial categories and inequalities are produced.
Takeaways & Limitations
Unclear or inconsistent racial conceptualizations and measurement practices can undermine the validity and utility of research results.
Abstract
from arXiv · showhide
We examine the way race and racial categories are adopted in algorithmic fairness frameworks. Current methodologies fail to adequately account for the socially constructed nature of race, instead adopting a conceptualization of race as a fixed attribute. Treating race as an attribute, rather than a structural, institutional, and relational phenomenon, can serve to minimize the structural aspects of algorithmic unfairness. In this work, we focus on the history of racial categories and turn to critical race theory and sociological work on race and ethnicity to ground conceptualizations of race for fairness research, drawing on lessons from public health, biomedical research, and social survey research. We argue that algorithmic fairness researchers need to take into account the multidimensionality of race, take seriously the processes of conceptualizing and operationalizing race, focus on social processes which produce racial inequality, and consider perspectives of those most affected by sociotechnical systems.
1 INTRODUCTION
Algorithmic fairness frameworks often encode social groups as formal dataset or algorithmic attributes, despite race and other protected categories being contextual, unstable, and politically constructed. The paper critiques fixed-attribute treatments of race and proposes a methodology attentive to its social construction, operationalization, multidimensionality, and affected communities.
- Fairness frameworks commonly formalize social groups in datasets or algorithms, although relevant groups are contextual and unstable social constructs.
- Race is a major axis of algorithmic resource allocation and representation, making race-based methodologies and categories important objects of critical evaluation.
- Current methodologies often treat race as a fixed attribute rather than a structural, institutional, and relational phenomenon, minimizing structural aspects of algorithmic unfairness.
- The paper treats racial measurement as a political project and calls for data collection and annotation grounded in the social and historical contexts of classification.
- Its approach reviews prior fairness research, racial-classification history, and disciplinary work on operationalization before arguing for multidimensional race, attention to racism, and affected perspectives.
2 THE PROBLEM WITH RACIAL CATEGORIES
The COMPAS debate illustrates how racial categories enter fairness analyses through institutional data whose classification practices may be unclear or historically shaped. The paper argues that replacing race with inferred race-like categories does not resolve the problem because race operates differently across systems of inequality and remains technologically and infrastructurally consequential.
- COMPAS fairness analyses relied on Broward County classifications resembling Census categories, but with omitted and redefined categories.
- The methodological appendix did not explain how defendant race was measured, leaving Broward County’s classification process unclear.
- Racial categories in algorithmic fairness research have largely gone unquestioned despite critiques of label instability and warnings against equating genetic ancestry with race or ethnicity.
- Using race-like categories?: Replacing existing racial categories with race-like categories inferred through unsupervised learning would still reify socially constructed race and sidestep the measurement problem.
- Using race-like categories?: Race-like categories depend on context because different systems of inequality require distinct modes of operationalization.
- Using race-like categories?: Categorization is infrastructural and requires examining who classifies, why classification occurs, and what broader ends it serves.
3 HISTORIES OF RACIAL CATEGORIZATION
Racial categories emerged through historical, political, institutional, and scientific processes rather than reflecting natural differences. Their classification and measurement continue to shape social groups, state practices, and the interpretation of racial disparities.
- Race is socially constructed through historical events, social forces, political power, and colonial conquest, while retaining material effects.
- Classification reflects the social, economic, and organizational imperatives of the institutions and occupations performing it.Categories can become infrastructure on which other structural and ideational elements are built.
- Racial classification in social statistics developed alongside nation-building, colonization, slavery, eugenics, and scientific racism.These histories linked statistical differentiation to projects of racial stratification and state power.
- States used racial boundaries in citizenship and administration, while censuses reduced complex populations to schematic categories for comparison and aggregation.Censuses also helped constitute the boundaries of racial social groups rather than merely counting them.
- Race measurement depends on the appraisal used, including self-identification, phenotype, and third-party observation, making race a constellation of multiple dimensions.Different dimensions can be measured differently and may produce differing empirical outcomes.
4 LESSONS FROM OTHER DISCIPLINES
Research across public health, biomedical science, and social inequality shows that racial variables are difficult to conceptualize and measure consistently. These lessons support treating race as multidimensional and socially situated, scrutinizing operationalization, and focusing on racism and structural conditions rather than race as a causal entity.
- Limitations in operationalizing race: Race is inconsistently conceptualized and measured across studies, undermining the validity and utility of findings and complicating international comparisons.Different classification schemes can mismatch measurements for the same person across studies or time points, affecting health, population, and inequality statistics.
- Limitations in operationalizing race: Standardizing racial taxonomies cannot resolve concerns about reifying race as a natural category or obscuring environmental, social, and structural contributors to disparities.Uncritical standardization may entrench racial stratification, misracialize diseases, and support biological essentialism.
- Limitations in operationalizing race: Misconstruing race as a causal variable can misattribute mechanisms to racial categories and produce ineffective public policy interventions.Critical objections include race's lack of manipulability and the risk of confusing racial stratification with race itself.
- Critical race methodologies: Critical race methodologies recommend denaturalizing race without dematerializing it by treating it as multidimensional, relational, and socially situated.They emphasize examining category choices, measurement schemes, and the justification of operationalization, while allowing context to determine whether racial categories are useful.
- Critical race methodologies: Public health research increasingly shifts from studying effects of race to studying racism and the institutional and structural conditions shaping racial disparities.This approach also emphasizes researcher positionality and affected communities' perspectives in research practice.
- Critical race methodologies: Descriptive racial analyses remain useful for identifying inequality, but category selection and assignment can substantially change their results.The appropriate operationalization depends on the outcome being studied, so descriptive use still requires explicit methodological justification.
5 IMPLICATIONS FOR USING RACE IN ALGORITHMIC FAIRNESS RESEARCH
The paper argues that algorithmic fairness research often reduces race to simplified, decontextualized variables, obscuring the historical, structural, and relational conditions that produce inequality. It recommends centering racial conceptualization and measurement, using context-sensitive analyses, and examining social processes beyond algorithmic outputs.
- Limits of group fairness: Race is often modeled as a single-dimensional variable, erasing the social, economic, and political complexity of racial categories.This simplification can also treat distinct groups as interchangeable.
- Limits of group fairness: Group fairness criteria abstract racial categories into mathematically comparable groups, obscuring their hierarchical and sociopolitical complexity.Equalizing statistics across groups can pursue sameness without addressing how groups are differently treated.
- Conceptualizing and operationalizing race: Race operationalization should be matched to the research purpose: observed race may suit discrimination studies, whereas self-identification may suit identity formation and voting behavior.The paper argues that measurement choices affect the scope and effectiveness of subsequent analysis and interventions.
- Conceptualizing and operationalizing race: Researchers should critically evaluate racial schemas because census and bureaucratic categories are unstable, contingent, and rooted in racial inequality.The paper centers conceptualizing and operationalizing race as part of fairness analysis.
- Conceptualizing and operationalizing race: Researchers should transparently report definitions, categories, measurements, and motivations, collect multiple race measures when possible, and treat measurement as an empirical problem.Survey research illustrates that racial category and measurement choices can affect measured outcomes.
- Conceptualizing and operationalizing race: Phenotypically defined categories may help analyze discrimination in observable-data settings, but they do not provide a complete solution and should not equate race with phenotype or biology.The paper warns that facial features can be repurposed to predict individual characteristics or psychological states, invoking histories of physiognomy and scientific racism.
- Limits of the algorithmic frame: Disaggregated analysis should begin pragmatically by asking why race matters to the system and how racial oppression is embedded in data, models, and resulting systems.The paper notes that severe fairness concerns may appear before quantitative race-based output analysis, including through racially biased source data.
6 CONCLUSION
The conclusion shifts attention from algorithms alone to the racial categories and classification histories embedded in fairness frameworks. It argues for multidimensional, historically grounded analysis and asks researchers to consider who creates classifications, whom they serve, and how they relate to accumulated injustice.
- Conclusion: The paper traces classification, eugenics, and state-sponsored category creation to challenge the naturalization of racial categories.It connects these histories to limitations in simplistic and decontextualized algorithmic fairness methodologies.
- Conclusion: Racial classifications should be interrogated by asking who categorizes, for what purpose, and who bears the classifications’ policy consequences.The paper retains post-hoc disaggregation as a possible use while stressing the need to contextualize categories and examine whom they serve.