Measurement and Fairness
Open accessThe DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index; 7 of the 70 sources its claims cite were checked for support.
Abstract
We propose measurement modeling from the quantitative social sciences as a framework for understanding fairness in computational systems. Computational systems often involve unobservable theoretical constructs, such as socioeconomic status, teacher effectiveness, and risk of recidivism. Such constructs cannot be measured directly and must instead be inferred from measurements of observable properties (and other unobservable theoretical constructs) thought to be related to them---i.e., operationalized via a measurement model. This process, which necessarily involves making assumptions, introduces the potential for mismatches between the theoretical understanding of the construct purported to be measured and its operationalization. We argue that many of the harms discussed in the literature on fairness in computational systems are direct results of such mismatches. We show how some of these harms could have been anticipated and, in some cases, mitigated if viewed through the lens of measurement modeling. To do this, we contribute fairness-oriented conceptualizations of construct reliability and construct validity that unite traditions from political science, education, and psychology and provide a set of tools for making explicit and testing assumptions about constructs and their operationalizations. We then turn to fairness itself, an essentially contested construct that has different theoretical understandings in different contexts. We argue that this contestedness underlies recent debates about fairness definitions: although these debates appear to be about different operationalizations, they are, in fact, debates about different theoretical understandings of fairness. We show how measurement modeling can provide a framework for getting to the core of these debates.
Overview
CoChat Summary
TL;DR
The paper proposes measurement modeling from the quantitative social sciences as a framework for understanding fairness in computational systems, arguing that many fairness harms stem from mismatches between unobservable theoretical constructs and their operationalizations. It contributes fairness-oriented construct reliability and construct validity tools and argues that debates over fairness definitions are really debates about different theoretical understandings of fairness—i.e., values.
Key findings
- Many fairness-related harms arise because computational systems operationalize unobservable constructs—such as socioeconomic status, teacher effectiveness, risk of recidivism, and patient benefit—via measurement models whose assumptions are left implicit or untested.
- The authors contribute fairness-oriented conceptualizations of construct reliability and construct validity, including seven aspects of construct validity: face, content, convergent, discriminant, predictive, hypothesis, and consequential validity.
- Recent fairness-definition debates—such as individual vs. group fairness and COMPAS’s error-rate balance vs. predictive parity—are framed as operationalization disputes but are actually debates about different theoretical understandings of fairness, and therefore about values.
- Examples include Obermeyer et al.’s finding that only 18% of patients identified for high-risk care management enrollment were Black; EVAAS teacher-effectiveness scores varying from 6/100 to 96/100 for one teacher; and the Universal Credit system treating £1000 paychecks on March 30 and April 29 as a £2000 monthly wage.
Methods
- Conceptual/position paper; no empirical data collection or new dataset.
- Uses five illustrative examples: measuring height, socioeconomic status, teacher effectiveness via EVAAS MRM, risk of recidivism via COMPAS, and patient benefit via high-risk care management models.
- Analysis is conceptual synthesis, uniting traditions from political science, education, and psychology to apply construct reliability and construct validity to fairness in computational systems.
Limitations
- Stated: the authors acknowledge that assessing construct reliability and construct validity can be time-consuming, and they omit inter-rater reliability and inter-item reliability from their conceptualization.
- Stated: they emphasize that values play a role in the measurement modeling process, not that measurement modeling is value-free.
- Evident: as a conceptual paper, it offers no new empirical validation or dataset; its examples and numbers rely on prior literature, documentation, and reported cases.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 13 Not valid · 48 Not checked · 116 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In contrast to computer scientists, who focus primarily on out-ofsample prediction, quantitative social scientists typically test their assumptions by assessing construct reliability and construct validity.”
§ Measurement and Fairness > 3 TESTING ASSUMPTIONS
What the sources say
- Partly supportedcites [49] matched to “How to analyze political attention with minimal assumptions and costs” (2010) sourceExperimental — accuracy not yet measured
The 'Reliability, Validity, Interpretation, and Application' section supports the substance of the claim: 'The evaluation of any measurement is generally based on its reliability (can it be repeated?) and validity (is it right?),' and the paper then assesses convergent and discriminant construct validity (with Weber 1990 cited) plus semantic, predictive and hypothesis validity. The companion 'Categorizing Texts: Methods, Assumptions, and Costs' section does describe supervised learning as evaluated by 'summaries of out-of-sample predictive accuracy,' which underwrites the contrasting half of the citing sentence. What our copy does not state is the disciplinary generalization itself: it never says that quantitative social scientists 'typically' test assumptions this way, nor that computer scientists focus primarily on out-of-sample prediction — the reliability/validity framing is presented as a general principle of measurement evaluation in content analysis, not as a characterization of a research community. (LLM, conf=0.70)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Prompted by this, many academic papers began to discuss the properties and shortcomings of different definitions of group fairness, with some papers demonstrating that it is impossible for a system like COMPAS to possess both error-rate balance and predictive parity when the probability of recidivism for Black defendants is not the same as the probability of recidivism for white defendants [9,11,1”
§ Measurement and Fairness > 4.4 Group Fairness
What the sources say
- Partly supportedcites [9] matched to “Fairness in Criminal Justice Risk Assessments: The State of the Art” (2021) sourceExperimental — accuracy not yet measured
Section 6.1 (p. 18) states that equal error rates and conditional use accuracy cannot both be achieved when groups have different base rates and classification is not perfect. The claim omits the paper’s qualification: with perfect separation, both can be achieved despite different base rates. (LLM, conf=0.88)
- Supportedcites [11] matched to “Fair prediction with disparate impact: a study of bias in recidivism prediction instruments” (2017) source
Section 2.1 (p. 2) says test fairness implies equal positive predictive value across groups, and that when recidivism prevalence differs, a test-fair score cannot have equal false-positive and false-negative rates across those groups. It applies this result to Black and White defendants assessed with COMPAS. (LLM, conf=0.93)
- Not checked yetcites [13]
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“by using arrests as a proxy for crimes committed, COMPAS can only exacerbate racist policing practices, rather than transcending them [7,13,23,37,39].”
§ Measurement and Fairness > 3.2.7 Consequential validity.
What the sources say
- Not checked yetcites [7]
- Not checked yetcites [13]
- Partly supportedcites [23] matched to “The false promise of risk assessments: epistemic reform and the limits of fairness” (2020) sourceExperimental — accuracy not yet measured
Section 5 says risk assessments use arrest data that typically provide a racially biased measure of crime, supporting the claim's concern about using arrests as a proxy. Sections 3 and 3.3 argue that these tools do not necessarily reduce racial discrimination and are poorly suited to structural reform. The paper does not establish the stronger claim that COMPAS can only exacerbate racist policing practices. (LLM, conf=0.84)
- Partly supportedcites [37] matched to “To Predict and Serve?” (2016) sourceExperimental — accuracy not yet measured
The ‘Bias in police-recorded data’ and ‘Discussion’ sections say that police records reflect unequal policing and that predictive policing can reproduce and magnify those biases. But the paper’s case study examines PredPol, not COMPAS; it does not establish that COMPAS uses arrests as a proxy for crimes committed or that it can only exacerbate racist policing. (LLM, conf=0.87)
- Not checked yetcites [39]
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Moreover, because most computational systems are developed by computer scientists, the practice of collapsing these distinctions is widespread [47].”
§ Measurement and Fairness > 5 DISCUSSION
What the sources say
- Partly supportedcites [47] matched to “Problem Formulation and Fairness” (2019) sourceExperimental — accuracy not yet measured
Our copy of Passi & Barocas (2019) does support the descriptive half of the claim: the Abstract states that 'the specification and operationalization of the problem are always negotiated and elastic, and rarely worked out with explicit normative considerations in mind,' and normative assessments 'often take them for granted'; the case-study section (4 CASE-STUDY: SPECIAL FINANCING) shows practitioners equating lead quality with financeability, financeability with dealer decisions, and dealer decisions with credit score ranges, i.e. collapsing a construct into its proxy/measure. What the paper nowhere says is the causal premise the citing sentence attaches to it — that this practice is widespread 'because most computational systems are developed by computer scientists'; no passage makes any claim about who develops most computational systems, and the BACKGROUND section instead stresses that applied data science is 'inherently collaborative,' with 'project managers, product designers, and business analysts ... as much a part of ... corporate data science as are data scientists' [36:16]. The paper's own explanation for such collapses is practical and organizational (business requirements, data availability, analytic cost), not the disciplinary composition of developers, so the attributing mechanism and the 'widespread' generalization beyond one ethnographed team are unsupported. (LLM, conf=0.70)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“We emphasize that even with careful consideration of different theoretical understandings and careful assessment of both construct reliability and construct validity, this approach is fraught and likely to cause a variety of fairness-related and other harms [e.g., 8, 22, 33, 35, 44], despite good intentions.”
§ Measurement and Fairness > 4.5 Demographic Factors
What the sources say
- Not checked yetcites [8] matched to “What is the point of fairness?” (2020) source
- Partly supportedcites [22] matched to “The Profiling Potential of Computer Vision and the Challenge of Computational Empiricism” (2019) sourceExperimental — accuracy not yet measured
Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
- Not checked yetcites [33] matched to “The Misgendering Machines: Trans/HCI Implications of Automatic Gender Recognition” (2018) source
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
77 more claims
Create a free account to see all 182 claims and to verify the 63 references still open.
- It may state two separate claims as one. 42
- Probably not a substantial claim. 82
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“When measurement bias is associated with a sensitive characteristic, in this case gender, the model is likely to replicate the pattern which can result in an unfair allocation of jobs [59].”
From Algorithmic Unfairness through the Lens of EU Non-Discrimination Law: Or Why the Law is not a Decision Tree (2023) · cites this paper as [59]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Essentially, equal calibration requires that the meaning of predicted scores is equal across groups [59]: receiving a score of 0.7 corresponds to a probability of 0.7, irrespective of sensitive group membership.”
From Algorithmic Unfairness through the Lens of EU Non-Discrimination Law: Or Why the Law is not a Decision Tree (2023) · cites this paper as [59]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“These evaluations should also demonstrate that the outcome of interest can be measured with reasonable accuracy and validity, as low-quality data and bad proxy variables present a significant limit on an algorithm's reliability ( Jacobs & Wallach, 2021 ).”
From The flaws of policies requiring human oversight of government algorithms (2022) · cites this paper as Jacobs & Wallach, 2021
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Framing bias as something to be measured objectively ignores the assumptions made in the operationalization of the measurement tool (Jacobs and Wallach 2021).”
From Bias and Fairness in Large Language Models: A Survey (2024) · cites this paper as Jacobs and Wallach 2021
Citations
Cites
Dwork, C., Hardt, M., Pitassi, T., Reingold, O., & Zemel, R. (2012) · Proceedings of the 3rd Innovations in Theoretical Computer Science Conference · doi:10.1145/2090236.2090255 · Cited by 3,501
Equality of Opportunity in Supervised Learning
Hardt, M., Price, E., & Srebro, N. (2016) · arXiv · doi:10.48550/arxiv.1610.02413 · Cited by 2,536 · cited 1×
“5Error-rate balance is also known as equalized odds [28].”
Fair prediction with disparate impact: a study of bias in recidivism prediction instruments
Chouldechova, A. (2017) · Big Data · doi:10.1089/big.2016.0047 · Cited by 2,159 · cited 2×
“r operationalization, i.e., calibration. The fairness in machine learning community responded, largely re ecting the incompatibilities and shortcomings of dierent fairness operationalizations (e.g., [8, 11, 15, 13, 39, 54]). Meanwhile, Chouldechova [11] and Kleinberg et al. [39] proved that the two prominent and desir…”
BackgroundInfluentialMachine Learning: An Applied Econometric Approach
Mullainathan, S., & Spiess, J. (2017) · Journal of Economic Perspectives · doi:10.1257/jep.31.2.87 · Cited by 1,989 · cited 1×
“Despite the name, predictive validity is not the same as out-of-sample prediction [29, 47].”
BackgroundFairness and Abstraction in Sociotechnical Systems
Selbst, A. D., Boyd, D., Friedler, S. A., Venkatasubramanian, S., & Vertesi, J. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287598 · Cited by 1,631 · cited 2×
“modeling brings the biggest value to fairness. In practice, evaluation of computational systems can fail to characterize potential harms which are otherwise captured by failures of construct validity [27, 53, 59]. The language of measurement, with the tools of construct validity and reliability, provide a concrete fra…”
BackgroundObjective Tests as Instruments of Psychological Theory
Loevinger, J. (1957) · Psychological Reports · doi:10.2466/pr0.1957.3.3.635 · Cited by 1,473
Algorithmic Decision Making and the Cost of Fairness
Corbett-Davies, S., Pierson, E., Feller, A., Goel, S., & Huq, A. (2017) · Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/3097983.3098095 · Cited by 1,149
Fairness in Criminal Justice Risk Assessments: The State of the Art
Berk, R., Heidari, H., Jabbari, S., Kearns, M., & Roth, A. (2021) · Sociological Methods & Research · doi:10.1177/0049124118782533 · Cited by 988 · cited 3×
“explicitly consider how to minimize or compensate for historical injustices [11]. This nontrivial measurement task is deeply complex and requires signicant debate about justice, policy, and fairness [8, 13, 15, 37]. 4.2 Value-Added Modeling in Education We now look to the well-known example of value-added modeling for…”
BackgroundInfluentialInherent Trade-Offs in the Fair Determination of Risk Scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016) · arXiv · doi:10.48550/arxiv.1609.05807 · Cited by 941
Social Data: Biases, Methodological Pitfalls, and Ethical Boundaries
Olteanu, A., Castillo, C., Diaz, F., & Kıcıman, E. (2019) · Frontiers in Big Data · doi:10.3389/fdata.2019.00013 · Cited by 763 · cited 2×
“, [27, 41, 50].”
Background
Research works printed but not linked: 17
- Audrey Amrein-Beardsley and Tray Geiger. 2020. Methodological Concerns About the Education Value-Added Assessment System (EVAAS): Validity, Relia- bility, and Bias. SAGE Open 10, 2 (2020), 2158244020922224.
- Julia Angwin, Jeff Larson, Surya Mattu, and Lauren Kirchner. 2016. Machine Bias. ProPublica (May 2016).
- Solon Barocas, Kate Crawford, Aaron Shapiro, and Hanna Wallach. 2017. The problem with bias: from allocative to representational harms in machine learning. Special Interest Group for Computing. Information and Society (SIGCIS) (2017).
- Geoffrey C Bowker and Susan Leigh Star. 2000. Sorting things out: Classification and its consequences. MIT Press. · doi:10.5555/380681
- Sam Corbett-Davies, Emma Pierson, Avi Feller, and Sharad Goel. 2016. A com- puter program used for bail and sentencing decisions was labeled biased against blacks. It's actually not that clear. Washington Post (October 2016).
- Kate Crawford. 2017. The Trouble with Bias. NIPS Keynote, https://www. youtube.com/watch?v=fMym_BKWQzk, accessed 08-22-2018.
- Walter Bryce Gallie. 1955. Essentially contested concepts. In Proceedings of the Aristotelian Society, Vol. 56. JSTOR, 167-198.
- Justin Grimmer. 2012. Comment: Evaluating model performance in fictitious prediction problems. Technical Report.
- Simon Jackman. 2008. Measurement. In The Oxford Handbook of Political Method- ology. Oxford Univ. Press, Chapter 9.
- Matthew Joseph, Michael Kearns, Jamie H Morgenstern, and Aaron Roth. 2016. Fairness in learning: Classic and contextual bandits. NIPS (2016).
- Sandra G Mayson. 2018. Bias in, bias out. Yale Law Journal 128 (2018), 2218.
- Samuel Messick. 1987. Validity. In ETS Research Report Series.
- Ziad Obermeyer, Brian Powers, Christine Vogeli, and Sendhil Mullainathan. 2019. Dissecting racial bias in an algorithm used to manage the health of populations. Science 366, 6464 (2019), 447-453.
- Cathy O'Neil. 2016. Weapons of Math Destruction. Broadway Books.
- Geoff Pleiss, Manish Raghavan, Felix Wu, Jon Kleinberg, and Kilian Q Weinberger. 2017. On fairness and calibration. NIPS (2017).
- Anna Roberts. 2018. Arrests as guilt. Ala. L. Rev. 70 (2018), 987.
- Human Rights Watch. 2020. UK: Automated Benefits System Failing People in Need. https://www.hrw.org/news/2020/09/29/uk-automated-benefits-system- failing-people-need
Cited by
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
PaLM: Scaling Language Modeling with Pathways
Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.02311 · Cited by 2,590
“…al., 2021b), there is a lack of standardization of fairness benchmarks, an understanding of what harms different bias measures in NLP relate to (Blodgett et al., 2020, 2021; Jacobs & Wallach, 2021), and coverage of identities in fluid, comprehensive ways (Cao & Daumé III, 2020; Dev et al., 2021a).”
BackgroundAnil, R., Dai, A. M., Firat, O., Johnson, M., Lepikhin, D., Passos, A., Shakeri, S., Taropa, E., Bailey, P., Chen, Z., et al. (2023) · arXiv · doi:10.48550/arxiv.2305.10403 · Cited by 834
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models
Gehman, S., Gururangan, S., Sap, M., Choi, Y., & Smith, N. A. (2020) · Findings of the Association for Computational Linguistics: EMNLP 2020 · doi:10.18653/v1/2020.findings-emnlp.301 · Cited by 693
Bias and Fairness in Large Language Models: A Survey
Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Kim, S., Dernoncourt, F., Yu, T., Zhang, R., & Ahmed, N. K. (2024) · Computational Linguistics · doi:10.1162/coli_a_00524 · Cited by 1,245
Metadata
- Authors
- Jacobs, A. Z., & Wallach, H.
- Affiliations
- University of Michigan (US)
- Microsoft (United States) (US)
- Venue
- Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency
- Publisher
- ACM
- Published
- 2021-03
- Pages
- 375-385
- OpenAlex
- W2995006168
- Semantic Scholar
- 681f4fbce872f138cbac9cdd92e8f6ed89ba6f8d
- Cited by
- 362
- References
- 24
- Citations by year
- 2026: 65 · 2025: 55 · 2024: 54 · 2023: 82 · 2022: 47 · 2021: 41
- FWCI
- 50.54
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- ACM Conference on Fairness, Accountability, and Transparency (FAccT) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Decision Sciences › Management Science and Operations Research › Psychometric Methodologies and Testing
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Keywords
- measurement modeling, fairness, construct reliability, construct validity, operationalization, fairness definitions, algorithmic fairness, socioeconomic status, teacher effectiveness, recidivism risk, essentially contested concepts, measurement assumptions
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Measurement and Fairness. https://app.cochat.ai/papers/10.1145/3442188.3445901
- The paper
- Jacobs, A. Z., & Wallach, H. (2021). Measurement and Fairness. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3442188.3445901
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.