Pith. sign in

REVIEW 4 major objections 7 minor 83 references

Using Machine Bias To Measure Human Bias

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A framework that trains one model per human decision-maker claims to recover that person's decision-bias gap from a small, disjoint set of gold-standard labels by dividing the model's adjusted true-positive-rate gap by a common ratio $c$.

desk verdict Real problem, plausible heuristic, but the theoretical guarantee is circular and the simulations favor the method; still worth a serious referee. read the letter →

arxiv 2411.18122 v4 pith:46BHMOR5 submitted 2024-11-27 cs.LG

Wanxue Dong , Maria De-Arteaga , Maytal Saar-Tsechansky This is my paper · ORCID
classification cs.LG
keywords humanlabelslabelbiasassessmentcrowdsourcingmachinelearningalgorithmicfairnesstruepositiverategapgoldstandardscarcity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes using machine bias as a measuring instrument for human bias. It trains a separate model on each decision-maker's past decisions, recalibrates the models so they share a common recall-versus-precision ratio $c$, and applies them to a small set of gold-standard labels that the decision-makers never saw. The central theoretical claim is that this yields the true gap in error rates across groups for each human. The method matters because current practice either compares selection rates, which confound base-rate differences with bias, or requires gold-standard labels for every assessed decision.

What carries the argument

The load-bearing object is the recall-versus-precision ratio $c=\mathrm{TPR}_{\hat Y|Y',A}/\mathrm{PPV}_{\hat Y|Y',A}$, which is also equal to the ratio of the number of positive predictions to the number of positive human decisions within each group. MDBA trains one base model per human, then adjusts group-specific classification thresholds until every model shares the same $c$; this post-processing step is designed to remove bias introduced during model training. The adjusted models are applied to the small gold-standard set to obtain $\mathrm{GAP}_{\hat Y|Y,A}$, and dividing by $c$ gives the human bias estimate. The lemma $\hat Y \perp Y \mid Y'$ is the algebraic hinge that turns this ratio identity into an exact equality.

What would settle it

Fit a correctly specified model to each of several decision-makers whose decisions and gold outcomes are both fully known, then check whether $\Pr(\hat Y=1 \mid Y'=1, A=a) = \Pr(\hat Y=1 \mid Y'=1, A=a, Y=1)$ for each group. Wherever this equality fails by a measurable amount, the identity $\mathrm{GAP}_{\hat Y|Y,A}/c = \mathrm{GAP}_{Y'|Y,A}$ will also fail, and the size of the failure shows how much of the method's estimate is model artifact rather than human bias.

Watch

Extended reading notes

Core claim

The central result, Theorem 2, states that when the functional form of a human's decision model is correct, there is a ratio $c=\mathrm{TPR}_{\hat Y|Y',A}/\mathrm{PPV}_{\hat Y|Y',A}$ such that $\mathrm{GAP}_{\hat Y|Y,A}/c = \mathrm{GAP}_{Y'|Y,A}$. In words, the disparity in true positive rates that the model exhibits between groups, divided by the model's over-prediction ratio, recovers the disparity the human decision-maker would exhibit if gold labels were available for every instance. The proof relies on a lemma that correct specification makes the model's prediction conditionally independent of the gold label given the human's decision. The experiments simulate biased decisions with and without correct internal ordering of candidates, on four datasets, and report that the proposed method has lower mean absolute error than the alternative benchmarks in most settings.

Load-bearing premise

The theory assumes that a correctly specified model of a person's decisions reproduces those decisions so faithfully that, knowing the person's decision, the gold-standard outcome adds no extra information about what the model would predict.

Editorial extensions

If this is right

  • Organizations could audit physicians, recruiters, or crowd workers using a few hundred gold-standard labels drawn from a disjoint sample, rather than labeling every decision.
  • Decision-makers who evaluated completely different instances can be compared, because each model is trained on that individual's own historical decisions and the gold-standard set only needs to come from the same distribution.
  • The estimate can be computed before and after an intervention, allowing organizations to measure whether a bias-mitigation policy actually changed the error-rate gap.
  • Because the method produces an estimate of bias in the human labels themselves, it can identify training sets that would propagate societal bias into downstream machine learning models.
  • The simulations show estimation error falling substantially as the gold-standard pool grows from 100 to 400 instances per group, so small investments in expert labels can yield large gains in measurement accuracy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The authors do not spell this out, but the same ratio-adjustment identity should transfer to other disparity metrics such as false-positive-rate gaps, since the proof only uses the group-conditional true positive rate and positive predictive value of the model relative to the human.
  • If the method works in practice, the fixed cost of a small expert panel becomes a reusable audit instrument, shifting the economics of bias measurement from per-decision labeling to one-time gold-standard collection.
  • A diagnostic the authors do not propose: the spread of $c$ values across fitted models is itself a warning signal, since a wide spread indicates that equalizing the ratio is forcing the models away from their natural behavior and the conditional-independence premise is likely violated.
  • The simulations intentionally violate the correct-functional-form assumption by generating gold labels with one model family and fitting human-decision models with another, so the natural next test is real human decisions with true gold labels, where the noise process is not generated by any known model.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes MDBA, a machine-learning framework for estimating human decision bias (defined as group disparities in true positive rates) when gold standard labels are scarce and possibly disjoint from the instances labeled by the humans under evaluation. The method trains one predictive model per human on that human's historical decisions, post-processes each model by recalibrating group-specific thresholds to equalize a recall-versus-precision ratio (RPR) across models and groups, applies the recalibrated models to the gold-standard set, and then divides the resulting model bias gap by the common RPR constant c to recover the human bias. The manuscript claims theoretical guarantees (Lemma, Theorem 1, Theorem 2) and reports extensive simulation experiments on four public datasets showing that MDBA outperforms selection rates, a gold-standard-based model, and confident learning.

Significance. The problem is important and timely: measuring bias in human decisions without full gold-standard labels has broad applications in healthcare, hiring, and crowdsourcing. The proposed pipeline is intuitively appealing, and the authors provide a reproducible simulation framework and code, which is a strength. However, the central theoretical claim is not sustained: the proof of the Lemma rests on an assertion that correct functional form implies the trained model is a deterministic copy of the human decision, which is false, and this invalidates the derived equality in Theorem 2. The empirical evaluation is explicitly conducted under conditions that violate the theoretical assumptions, so the experiments do not validate the theorems and instead function as evidence for a heuristic. If the theoretical claims were corrected or appropriately weakened, the empirical framework could serve as a useful benchmark for future work, but as stated the paper's core contribution is not supported.

major comments (4)
  1. [Appendix A, Theorem 1] The proof asserts 'Given the correct functional form for fk : Xk -> Y'_k then Yhat_k = Y'_k + epsilon' and from this concludes Yhat_k ⊥ Y_k | Y'_k. Correct functional form means the chosen model class contains the true decision function; it does not imply that the estimated model is a deterministic copy of the human decision, nor that any noise is constant. In general Yhat is a function of X estimated from finite data, and Y' may contain human noise, so Yhat and Y can be dependent even conditional on Y'. This conditional independence is load-bearing: equations (13), (14), and (18) in the proof of Theorem 1, and consequently equation (20) in Theorem 2, all rely on it. Without a valid proof of the Lemma, the theoretical guarantee is not established.
  2. [Section 4.2, Theorem 2] The proof assumes that there exists a single constant c such that for every human model and both protected groups the ratio TPR_{Yhat|Y',A}/PPV_{Yhat|Y',A} can be set to c by threshold recalibration. This existence is asserted, not proven. If no threshold yields a common c for all models and groups, Algorithm 1 cannot be executed. If multiple thresholds do, then different c values give different estimates of GAP_{Y'|Y,A} because the estimate is GAP_{Yhat|Y,A}/c; the suggestion to average over thresholds does not follow from the theorem, which guarantees equality only for the specific c satisfying equation (16), not for an average of several c's.
  3. [Section 5, paragraph beginning 'It is crucial to highlight'] Dividing the model gap by c rescales by a quantity that is the model's own recall-precision ratio with respect to Y'. Since c is determined by the same thresholding procedure that the method uses to equalize RPR across groups, the claimed recovery of GAP_{Y'|Y,A} is essentially a rearrangement of the model's confusion w.r.t. Y' rather than an independent measurement. Equation (18) shows that c directly multiplies the human TPR terms; any error or arbitrariness in c scales the estimated human bias linearly. The paper provides no error analysis or sensitivity analysis for the choice of c, which is central to the method's output.
  4. [Section 5] The empirical evaluation is deliberately constructed to violate the theoretical assumptions ('we purposefully create all simulations such that the core assumption underlying the theoretical guarantees of our method is violated'). This is a strength for robustness, but it also means the reported MAE improvements cannot be used to support the theoretical claims in Section 4.2; the experiments test a different regime in which the Lemma is known to fail. The paper should either restrict its claims to an empirical heuristic or redesign the experiments to also validate the theory under settings where the assumptions hold.
minor comments (7)
  1. [Appendix B] The text 'PRP ratio = c' appears to be a typo for 'RPR ratio = c'.
  2. The heading 'theoretical garantee removed' contains a typo: 'garantee' should be 'guarantee'.
  3. The word 'advencing' in the final sentence should be 'advancing'.
  4. The phrases 'methodoloigies' and 'decison makers' are typos and should be corrected.
  5. 'refereed as GS-based' should be 'referred to as GS-based'.
  6. The text reports '95% confidence bound' while the body text in Section 6 reports '90% confidence bounds'; these should be reconciled.
  7. The notation 'Yhat_kY_k|Y'_k' in the proof appears to be missing the independence symbol; it should be 'Yhat_k ⊥ Y_k | Y'_k'.

Circularity Check

2 steps flagged · score 6.0 of 10

Theoretical guarantee reduces to assuming the trained model is a copy of the human; empirical benchmark remains independent.

  1. self definitional [Appendix A, Lemma (proof)]
    "Given the correct functional form for fk : Xk 7→ Y′ k then ˆYk = Yk′ + ϵ where the ϵ is the constant term; and thus, ˆYkYk|Yk′."

    The Lemma assumes that 'correct functional form specification' means the trained model's predictions equal the human decisions up to a constant: Yhat = Y' + epsilon. That is not a consequence of correct specification; it is a definitional choice that builds the target quantity (human decision bias) into the model. The conditional independence Yhat ⊥ Y | Y' derived here is precisely the condition under which the model's error gap with respect to Y is a rescaled copy of the human's error gap, so Theorem 2's conclusion is assumed by the Lemma rather than derived.

  2. fitted input called prediction [Section 4.2, Theorem 2 and proof in Appendix A]
    "From Lemma and Theorem 1, we have P( ˆYi=1|Y′ i=1,A=a) P(Y′ i=1| ˆYi=1,A=a) = T PRi ˆY|Y′, A PPV i ˆY|Y′, A = c (16) Therefore, GAP ˆY|Y, A c = GAP Y′|Y, A (20)"

    The conversion factor c is not an externally measured quantity: it is the TPR/PPV ratio of the same fitted model with respect to the human labels Y' (eq. 16-17). Theorem 2's equality P(Yhat|A,Y)/c = P(Y'|A,Y) is obtained by substituting the Lemma's conditional-independence assumption, which already encodes that the model replicates the human decision function. Hence the 'measurement' of human bias is a rescaling of the fitted model's own bias by a factor computed from the same fitted model; the claimed prediction reduces to the input assumption Yhat = Y' + epsilon.

full rationale

The paper's theoretical derivation is circular at its root. The Lemma in Appendix A asserts that correct functional form implies Yhat = Y' + epsilon, which is not a consequence of correct specification but an assumption that the machine-learning model is a deterministic copy of the human decision process up to a constant. From that assumption, conditional independence Yhat ⊥ Y | Y' follows, and then Theorems 1 and 2 algebraically convert the model's TPR gap into the human's TPR gap by dividing by c, where c is the same model's recall-precision ratio with respect to the human labels. The central 'theoretical guarantee' therefore reduces by construction to the input assumption. However, the paper's empirical evaluation is self-contained and not circular: the simulations deliberately violate the functional-form assumption (gold labels generated with logistic regression, human-decision models fit with XGBoost), and MDBA is compared against SR, GS-based, and Confident Learning with reported MAE and significance tests. Those benchmark results provide independent evidence for the method's practical performance even though the stated theoretical guarantee does not. The score of 6 reflects that one or more of the paper's central predictions reduce by construction, while the empirical contribution stands independently.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central derivation depends on one ad hoc assumption that makes the model a deterministic copy of the human decision, plus an existence assumption for the threshold ratio c. No new physical or conceptual entities are introduced.

free parameters (2)
  • c (RPR ratio target) = not reported; varies with experiment
    Input to Algorithm 1 that controls threshold recalibration. The paper asserts existence but gives no selection rule; the final bias estimate is divided by c, so the output depends on this arbitrary parameter.
  • Group-specific thresholds pi^k_{A=a} and pi^k_{A=not a} = chosen to satisfy RPR = c
    Recalibrated per model and per group; the search procedure is not described in detail, and multiple solutions are averaged without a defined rule.
assumptions (4)
  • ad hoc to paper Correct functional form implies Yhat = Y' + epsilon, and hence Yhat independent of Y given Y', and Y' independent of Y given Yhat.
    Used in Appendix A Lemma and Theorem proofs; not a standard result and false for typical noisy human decisions. It makes the model a deterministic copy of the human decision, so machine bias equals human bias by construction.
  • ad hoc to paper There exists a single value c such that every human model can be thresholded to achieve TPR/PPV = c for both protected groups.
    Theorem 1 asserts existence but offers no constructive procedure or conditions under which thresholds exist; Algorithm 1 takes c as an input, implying existence is assumed for all datasets.
  • domain assumption The small gold-standard set GS is sampled i.i.d. from the same distribution as the humans' instance sets, and gold-standard labels are unbiased (reflect organizational goals).
    Used in Section 3 Problem Formulation; needed for estimates on GS to generalize to the humans' decision sets.
  • domain assumption The sensitive attribute A is binary and observed for all instances.
    Used throughout, e.g., Equation 1 and Algorithm 1; the method is defined only for A in {a, not a}.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Using Machine Bias To Measure Human Bias." pith.science (2026). https://pith.science/paper/46BHMOR5

@misc{pith2026241118122,
  author       = {Pith},
  title        = {Pith review of: Using Machine Bias To Measure Human Bias},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/46BHMOR5}},
  note         = {Machine review of arXiv:2411.18122}
}
read the original abstract

Biased human decisions have consequential impacts across various domains, yielding unfair treatment of individuals and resulting in suboptimal outcomes for organizations and society. In recognition of this fact, organizations regularly design and deploy interventions aimed at mitigating these biases. However, measuring human decision biases remains an important but elusive task. Organizations are frequently concerned with mistaken decisions disproportionately affecting one group. In practice, however, this is typically not possible to assess due to the scarcity of a gold standard: a label that indicates what the correct decision would have been. In this work, we propose a machine learning-based framework to assess bias in human-generated decisions when gold standard labels are scarce. We provide theoretical guarantees and empirical evidence demonstrating the superiority of our method over existing alternatives. This proposed methodology establishes a foundation for transparency in human decision-making, carrying substantial implications for managerial duties, and offering potential for alleviating algorithmic biases when human decisions are used as labels to train algorithms.

Figures

Figures reproduced from arXiv: 2411.18122 by the authors.

Figure 1
Figure 1. An illustration of K human decision-makers’ decision instance sets (left), S = {S Hk } K k=1 where S Hk = {X k i ,A k i ,Y ′k i } n k i=1 and a small non-overlapping set with gold-standard labels (right), GS = {Xl ,Al ,Yl} m l=1 . We seek to produce assessment of humans’ decision biases. We define bias as disparities in rates of error for a certain type of error of interest ([17]), such as gaps in true positive rate… view at source ↗
Figure 2
Figure 2. Key Steps of the MDBA methodt 4.2 Theoretical Analysis We now show that, given the correct functional form specification of the humans’ models, i.e. assuming that the functional form of the relationship between the dependent variables and the human decisions, f : X 7→ Y ′ , is correctly specified, our method can accurately assess humans’ decision bias. We demonstrate this in the series of theorems below. The proofs … view at source ↗
Figure 4
Figure 4. MDBA’s performance relative to benchmarks under scenarios where humans exhibit incorrect within-group orderings and datasets are with 20% and 30% positive label prevalence. MAE measure for humans’ decision bias estimation errors for our approach MDBA and the baseline approaches. Results are reported given a varying number of gold standard (GS) instances, different datasets with different positive label prevalence. T… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

83 extracted references · 71 canonical work pages

  1. [1]

    Unequal treatment: confronting racial and ethnic disparities in health care

    Alan Nelson. Unequal treatment: confronting racial and ethnic disparities in health care. Journal of the national medical association, 94(8):666, 2002

    2002

  2. [2]

    The social psychology of inequality

    Jolanda Jetten and Kim Peters. The social psychology of inequality. Springer, 2019

    2019

  3. [3]

    The language of social inequality

    Martha Augoustinos and Peta Callaghan. The language of social inequality. The social psychology of inequality, pages 321–334, 2019

    2019

  4. [4]

    Are emily and greg more employable than lakisha and jamal? a field experiment on labor market discrimination

    Marianne Bertrand and Sendhil Mullainathan. Are emily and greg more employable than lakisha and jamal? a field experiment on labor market discrimination. American economic review, 94(4):991–1013, 2004

    work page 2004

  5. [5]

    Expecting the unexpected: Effects of data collection design choices on the quality of crowdsourced user-generated content

    Roman Lukyanenko, Je ffrey Parsons, Yolanda F Wiersma, and Mahed Maddah. Expecting the unexpected: Effects of data collection design choices on the quality of crowdsourced user-generated content. MIS Quarterly, 43(2):623–647, 2019

    work page 2019

  6. [6]

    Real-e ffort incentives in online labor markets: Punishments and rewards for individuals and groups

    Matthew J Hashim and Jesse C Bockstedt. Real-e ffort incentives in online labor markets: Punishments and rewards for individuals and groups. MIS Quarterly, 48(1), 2024

    work page 2024

  7. [7]

    Data collection in the digital age

    Zachary R Steelman, Bryan I Hammer, and Moez Limayem. Data collection in the digital age. MIS quarterly, 38(2):355–378, 2014

    work page 2014

  8. [8]

    Cost-e ffective quality assurance in crowd labeling

    Jing Wang, Panagiotis G Ipeirotis, and Foster Provost. Cost-e ffective quality assurance in crowd labeling. Information Systems Research, 28(1):137–158, 2017

    work page 2017

Show all 83 references
  1. [9]

    Hate speech classifiers learn normative social stereotypes

    Aida Mostafazadeh Davani, Mohammad Atari, Brendan Kennedy, and Morteza Dehghani. Hate speech classifiers learn normative social stereotypes. Transactions of the Association for Computational Linguistics, 11:300–319, 2023

    2023

  2. [10]

    Mohammed Saeed, Nicolas Traub, Maelle Nicolas, Gianluca Demartini, and Paolo Papotti. Crowdsourced fact-checking at twitter: How does the crowd compare with experts? In Proceedings of the 31st ACM International Conference on Information & Knowledge Management, pages 1736–1746, 2022

    2022

  3. [11]

    Ai as an apolitical referee: Using alternative sources to decrease partisan biases in the processing of fact-checking messages

    Myojung Chung, Won-Ki Moon, and S Mo Jones-Jang. Ai as an apolitical referee: Using alternative sources to decrease partisan biases in the processing of fact-checking messages. Digital Journalism, pages 1–22, 2023

    2023

  4. [12]

    Big data’s disparate impact

    Solon Barocas and Andrew D Selbst. Big data’s disparate impact. Calif. L. Rev., 104:671, 2016

    2016

  5. [13]

    Tackling unconscious bias in hiring practices: The plight of the rooney rule

    Brian W Collins. Tackling unconscious bias in hiring practices: The plight of the rooney rule. NYUL Rev., 82:870, 2007

    2007

  6. [14]

    Facebook is testing the ‘rooney rule’approach to hiring

    Deepa Seetharaman. Facebook is testing the ‘rooney rule’approach to hiring. The Wall Street Journal, 2015

    2015

  7. [15]

    Fair prediction with disparate impact: A study of bias in recidivism prediction instruments

    Alexandra Chouldechova. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big data, 5(2):153–163, 2017

    2017

  8. [16]

    Algorithmic fairness in business analytics: Directions for research and practice

    Maria De-Arteaga, Stefan Feuerriegel, and Maytal Saar-Tsechansky. Algorithmic fairness in business analytics: Directions for research and practice. Production and Operations Management, 31(10):3749–3770, 2022

    2022

  9. [17]

    Equality of opportunity in supervised learning

    Moritz Hardt, Eric Price, and Nati Srebro. Equality of opportunity in supervised learning. Advances in neural information processing systems, 29:3315–3323, 2016

    2016

  10. [18]

    Evaluation of uterine cervix segmentations using ground truth from multiple experts

    Shiri Gordon, Shelly Lotenberg, Rodney Long, Sameer Antani, Jose Jeronimo, and Hayit Greenspan. Evaluation of uterine cervix segmentations using ground truth from multiple experts. Computerized Medical Imaging and Graphics, 33(3):205–216, 2009

    2009

  11. [19]

    Scaling up fact-checking using the wisdom of crowds

    Jennifer Allen, Antonio A Arechar, Gordon Pennycook, and David G Rand. Scaling up fact-checking using the wisdom of crowds. Preprint at https://doi. org/10.31234/osf. io/9qdza, 2020

    2020 doi

  12. [20]

    Racial-ethnic biases, time pressure, and medical decisions

    Irena Stepanikova. Racial-ethnic biases, time pressure, and medical decisions. Journal of health and social behavior, 53(3):329–343, 2012

    2012

  13. [21]

    Impact of race on cardiac care and outcomes in veterans with acute myocardial infarction

    Laura A Petersen, Steven M Wright, Eric D Peterson, and Jennifer Daley. Impact of race on cardiac care and outcomes in veterans with acute myocardial infarction. Medical care, pages I86–I96, 2002

    2002

  14. [22]

    We are the people and you are fake news: A social identity approach to populist citizens’ false consensus and hostile media perceptions

    Anne Schulz, Werner Wirth, and Philipp Müller. We are the people and you are fake news: A social identity approach to populist citizens’ false consensus and hostile media perceptions. Communication research, 47(2):201–226, 2020

    2020

  15. [23]

    Social-cognitive theory in encyclopedia of psychology (vol

    A Bandura and AE Kazdin. Social-cognitive theory in encyclopedia of psychology (vol. 7). Washington, DC, New York, NY: American Psychological Association.[Google Scholar], 2000. 19 A preprint - December 11, 2024

    2000

  16. [24]

    Physicians and implicit bias: how doctors may unwittingly perpetuate health care disparities

    Elizabeth N Chapman, Anna Kaatz, and Molly Carnes. Physicians and implicit bias: how doctors may unwittingly perpetuate health care disparities. Journal of general internal medicine, 28:1504–1510, 2013

    2013

  17. [25]

    Implicit bias among physicians and its prediction of thrombolysis decisions for black and white patients

    Alexander R Green, Dana R Carney, Daniel J Pallin, Long H Ngo, Kristal L Raymond, Lisa I Iezzoni, and Mahzarin R Banaji. Implicit bias among physicians and its prediction of thrombolysis decisions for black and white patients. Journal of general internal medicine, 22:1231–1238, 2007

    2007

  18. [26]

    Editorial for the special section on humans, algorithms, and augmented intelligence: The future of work, organizations, and society

    Hemant Jain, Balaji Padmanabhan, Paul A Pavlou, and TS Raghu. Editorial for the special section on humans, algorithms, and augmented intelligence: The future of work, organizations, and society. Information Systems Research, 32(3):675–687, 2021

    2021

  19. [27]

    Algorithmic bias: review, synthesis, and future research directions

    Nima Kordzadeh and Maryam Ghasemaghaei. Algorithmic bias: review, synthesis, and future research directions. European Journal of Information Systems, 31(3):388–409, 2022

    2022

  20. [28]

    Big data analytics capabilities: a systematic literature review and research agenda

    Patrick Mikalef, Ilias O Pappas, John Krogstie, and Michail Giannakos. Big data analytics capabilities: a systematic literature review and research agenda. Information systems and e-business management, 16:547–578, 2018

    2018

  21. [29]

    Business intelligence and analytics: From big data to big impact

    Hsinchun Chen, Roger HL Chiang, and Veda C Storey. Business intelligence and analytics: From big data to big impact. MIS quarterly, pages 1165–1188, 2012

    2012

  22. [30]

    Designing ethical algorithms

    Kirsten Martin. Designing ethical algorithms. MIS Quarterly Executive June, 2019

    2019

  23. [31]

    Artificial intelligence in information systems: State of the art and research roadmap

    Pär J Ågerfalk, Kieran Conboy, Kevin Crowston, J Eriksson Lundström, Sirkka L Jarvenpaa, Sudha Ram, and Patrick Mikalef. Artificial intelligence in information systems: State of the art and research roadmap. In Association for Information Systems, 2022

    2022

  24. [32]

    Fairness in decision-making—the causal explanation formula

    Junzhe Zhang and Elias Bareinboim. Fairness in decision-making—the causal explanation formula. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018

    2018

  25. [33]

    Evaluating the predictive validity of the compas risk and needs assessment system

    Tim Brennan, William Dieterich, and Beate Ehret. Evaluating the predictive validity of the compas risk and needs assessment system. Criminal Justice and behavior, 36(1):21–40, 2009

    2009

  26. [34]

    Method and system for loan origination and underwriting, 2007

    John F Mahoney and James M Mohen. Method and system for loan origination and underwriting, 2007

    2007

  27. [35]

    Machine bias

    Julia Angwin, Je ff Larson, Surya Mattu, and Lauren Kirchner. Machine bias. In Ethics of data and analytics, pages 254–264. Auerbach Publications, 2022

    2022

  28. [36]

    Harini Suresh and John V . Guttag. A framework for understanding unintended consequences of machine learning. CoRR, abs/1901.10002, 2019

    1901 arXiv

  29. [37]

    Ai: The issue of bias

    Vincent Violago and Nikko Quevada. Ai: The issue of bias. Managing Intell. Prop., 277:32, 2018

    2018

  30. [38]

    Amazon scraps secret ai recruiting tool that showed bias against women

    Je ffrey Dastin. Amazon scraps secret ai recruiting tool that showed bias against women. In Ethics of Data and Analytics, pages 296–299. Auerbach Publications, 2018

    2018

  31. [39]

    An overview of ethical issues in using ai systems in hiring with a case study of amazon’s ai based hiring tool

    Akhil Alfons Kodiyan. An overview of ethical issues in using ai systems in hiring with a case study of amazon’s ai based hiring tool. Researchgate Preprint, pages 1–19, 2019

    2019

  32. [40]

    When more data lead us astray: Active data acquisition in the presence of label bias

    Yunyi Li, Maria De-Arteaga, and Maytal Saar-Tsechansky. When more data lead us astray: Active data acquisition in the presence of label bias. In Proceedings of the AAAI Conference on Human Computation and Crowdsourcing, volume 10, pages 133–146, 2022

    2022

  33. [41]

    Editor’s comments: information systems research and behavioral economics

    Paulo B Goes. Editor’s comments: information systems research and behavioral economics. MIS quarterly, 37(3):iii–viii, 2013

    2013

  34. [42]

    Gender and ethnicity bias in medicine: A text analysis of 1.8 million critical care records

    David M Markowitz. Gender and ethnicity bias in medicine: A text analysis of 1.8 million critical care records. PNAS nexus, 1(4):pgac157, 2022

    2022

  35. [43]

    The colorblind crowd? founder race and performance in crowdfunding

    Peter Younkin and Venkat Kuppuswamy. The colorblind crowd? founder race and performance in crowdfunding. Management Science, 64(7):3269–3287, 2018

    2018

  36. [44]

    How unbecoming of you: Online experiments uncovering gender biases in perceptions of ridesharing performance

    Brad Greenwood, Idris Adjerid, Corey M Angst, and Nathan L Meikle. How unbecoming of you: Online experiments uncovering gender biases in perceptions of ridesharing performance. Journal of Business Ethics, pages 1–20, 2020

    2020

  37. [45]

    Social media and selection: Political issue similarity, liking, and the moderating effect of social media platform

    Julie T Wade, Philip L Roth, Jason Bennett Thatcher, and Michael Dinger. Social media and selection: Political issue similarity, liking, and the moderating effect of social media platform. MIS Quarterly, 44(3), 2020

    2020

  38. [46]

    Human decision-making biases in the moral dilemmas of autonomous vehicles

    Darius-Aurel Frank, Polymeros Chrysochou, Panagiotis Mitkidis, and Dan Ariely. Human decision-making biases in the moral dilemmas of autonomous vehicles. Scientific reports, 9(1):13080, 2019

    2019

  39. [47]

    Failures of fairness in automation require a deeper understanding of human-ml augmentation

    Mike HM Teodorescu, Lily Morse, Yazeed Awwad, and Gerald C Kane. Failures of fairness in automation require a deeper understanding of human-ml augmentation. MIS quarterly, 45(3), 2021. 20 A preprint - December 11, 2024

    2021

  40. [48]

    Do gender and race make a difference in acute coronary syndrome pretest probabilities in the emergency department? Academic Emergency Medicine, 24(2):142–151, 2017

    Paul I Musey Jr and Je ffrey A Kline. Do gender and race make a difference in acute coronary syndrome pretest probabilities in the emergency department? Academic Emergency Medicine, 24(2):142–151, 2017

    2017

  41. [49]

    Disparities in the emergency department evaluation of chest pain patients

    Liliana E Pezzin, Penelope M Keyl, and Gary B Green. Disparities in the emergency department evaluation of chest pain patients. Academic Emergency Medicine, 14(2):149–156, 2007

    2007

  42. [50]

    Gender bias in cardiovascular testing persists after adjustment for presenting characteristics and cardiac risk

    Anna Marie Chang, Bryn Mumma, Keara L Sease, Jennifer L Robey, Frances S Shofer, and Judd E Hollander. Gender bias in cardiovascular testing persists after adjustment for presenting characteristics and cardiac risk. Academic Emergency Medicine, 14(7):599–605, 2007

    2007

  43. [51]

    Organizational experiences and career success of mis professionals and managers: An examination of race differences

    Magid Igbaria and Wayne M Wormley. Organizational experiences and career success of mis professionals and managers: An examination of race differences. MIS Quarterly, pages 507–529, 1992

    1992

  44. [52]

    V ocal minority and silent majority: How do online ratings reflect population perceptions of quality

    Gao Guodong, Greenwood Brad, Agarwal Ritu, and McCullough Je ffrey. V ocal minority and silent majority: How do online ratings reflect population perceptions of quality. MIS Quarterly, 39(3):565–89, 2015

    2015

  45. [53]

    Toward a social psychology of race and race relations for the twenty-first century

    Jennifer A Richeson and Samuel R Sommers. Toward a social psychology of race and race relations for the twenty-first century. Annual review of psychology, 67:439–463, 2016

    2016

  46. [54]

    On the interplay of data and cognitive bias in crisis information management: An exploratory study on epidemic response

    David Paulus, Ramian Fathi, Frank Fiedrich, Bartel Van de Walle, and Tina Comes. On the interplay of data and cognitive bias in crisis information management: An exploratory study on epidemic response. Information Systems Frontiers, 26(2):391–415, 2024

    2024

  47. [55]

    On self-selection biases in online product reviews

    Nan Hu, Paul A Pavlou, and Jie Zhang. On self-selection biases in online product reviews. MIS quarterly, 41(2):449–475, 2017

    2017

  48. [56]

    Who are you and what are you selling? creatorbased and product-based racial cues in crowdfunding

    Lauren Rhue and Jessica Clark. Who are you and what are you selling? creatorbased and product-based racial cues in crowdfunding. Mis Quarterly, 46(4), 2022

    2022

  49. [57]

    A systematic review of the impact of physician implicit racial bias on clinical decision making

    Erin Dehon, Nicole Weiss, Jonathan Jones, Whitney Faulconer, Elizabeth Hinton, and Sarah Sterling. A systematic review of the impact of physician implicit racial bias on clinical decision making. Academic Emergency Medicine, 24(8):895–904, 2017

    2017

  50. [58]

    Forming impressions from stereotypes, traits, and behaviors: A parallel-constraint-satisfaction theory

    Ziva Kunda and Paul Thagard. Forming impressions from stereotypes, traits, and behaviors: A parallel-constraint-satisfaction theory. Psychological review, 103(2):284, 1996

    1996

  51. [59]

    Pomdp-based control of workflows for crowdsourcing

    Peng Dai, Christopher H Lin, Daniel S Weld, et al. Pomdp-based control of workflows for crowdsourcing. Artificial Intelligence, 202:52–85, 2013

    2013

  52. [60]

    Aggregating crowdsourced binary ratings

    Nilesh Dalvi, Anirban Dasgupta, Ravi Kumar, and Vibhor Rastogi. Aggregating crowdsourced binary ratings. In Proceedings of the 22nd international conference on World Wide Web, pages 285–294, 2013

    2013

  53. [61]

    Estimating diagnostic accuracy of raters without a gold standard by exploiting a group of experts

    Bo Zhang, Zhen Chen, and Paul S Albert. Estimating diagnostic accuracy of raters without a gold standard by exploiting a group of experts. Biometrics, 68(4):1294–1302, 2012

    2012

  54. [62]

    Maximum likelihood estimation of observer error-rates using the em algorithm

    Alexander Philip Dawid and Allan M Skene. Maximum likelihood estimation of observer error-rates using the em algorithm. Journal of the Royal Statistical Society: Series C (Applied Statistics), 28(1):20–28, 1979

    1979

  55. [63]

    Majority voting and pairing with multiple noisy labeling

    Victor Sheng, Jing Zhang, Bin Gu, and Xindong Wu. Majority voting and pairing with multiple noisy labeling. IEEE Transactions on Knowledge and Data Engineering, 2017

    2017

  56. [64]

    Learning from noisy labels by regularized estimation of annotator confusion

    Ryutaro Tanno, Ardavan Saeedi, Swami Sankaranarayanan, Daniel C Alexander, and Nathan Silberman. Learning from noisy labels by regularized estimation of annotator confusion. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 11244–11253, 2019

    2019

  57. [65]

    Confident learning: Estimating uncertainty in dataset labels

    Curtis Northcutt, Lu Jiang, and Isaac Chuang. Confident learning: Estimating uncertainty in dataset labels. Journal of Artificial Intelligence Research, 70:1373–1411, 2021

    2021

  58. [66]

    Dissecting racial bias in an algorithm used to manage the health of populations

    Ziad Obermeyer, Brian Powers, Christine V ogeli, and Sendhil Mullainathan. Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464):447–453, 2019

    2019

  59. [67]

    Man is to computer programmer as woman is to homemaker? debiasing word embeddings

    Tolga Bolukbasi, Kai-Wei Chang, James Zou, Venkatesh Saligrama, and Adam Kalai. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Proceedings of the 30th International Conference on Neural Information Processing Systems, NIPS’16, page 4356–4...

    2016

  60. [68]

    Digital first: The ontological reversal and new challenges for is research

    Richard L Baskerville, Michael D Myers, and Youngjin Yoo. Digital first: The ontological reversal and new challenges for is research. MIS quarterly, 44(2):509–523, 2020

    2020

  61. [69]

    Managing artificial intelligence

    Nicholas Berente, Bin Gu, Jan Recker, and Radhika Santhanam. Managing artificial intelligence. MIS quarterly, 45(3), 2021. 21 A preprint - December 11, 2024

    2021

  62. [70]

    Questioning racial and gender bias in ai-based recommendations: Do espoused national cultural values matter? Information Systems Frontiers, 24(5):1465–1481, 2022

    Manjul Gupta, Carlos M Parra, and Denis Dennehy. Questioning racial and gender bias in ai-based recommendations: Do espoused national cultural values matter? Information Systems Frontiers, 24(5):1465–1481, 2022

    2022

  63. [71]

    Editor’s comments: Computationally intensive theory construction: A primer for authors and reviewers

    Shalia Miranda, Nicholas Berente, Stefan Seidel, Hani Safadi, and Andrew Burton-Jones. Editor’s comments: Computationally intensive theory construction: A primer for authors and reviewers. MIS quarterly, 46(2):iii–xviii, 2022

    2022

  64. [72]

    Unifying the role of it in hyperturbulence and competitive advantage via a multilevel perspective of is strategy

    Ning Nan and Hüseyin Tanriverdi. Unifying the role of it in hyperturbulence and competitive advantage via a multilevel perspective of is strategy. Mis Quarterly, 41(3):937–A8, 2017

    2017

  65. [73]

    Who is a better decision maker? data-driven expert ranking under unobserved quality

    Tomer Geva and Maytal Saar-Tsechansky. Who is a better decision maker? data-driven expert ranking under unobserved quality. Production and operations management, 30(1):127–144, 2021

    2021

  66. [74]

    Repeated labeling using multiple noisy labelers

    Panagiotis G Ipeirotis, Foster Provost, Victor S Sheng, and Jing Wang. Repeated labeling using multiple noisy labelers. Data Mining and Knowledge Discovery, 28:402–441, 2014

    2014

  67. [75]

    Learning from crowds

    Vikas C Raykar, Shipeng Yu, Linda H Zhao, Gerardo Hermosillo Valadez, Charles Florin, Luca Bogoni, and Linda Moy. Learning from crowds. Journal of Machine Learning Research, 11(4), 2010

    2010

  68. [76]

    The right music at the right time: Adaptive personalized playlists based on sequence modeling

    Elad Liebman, Maytal Saar-Tsechansky, and Peter Stone. The right music at the right time: Adaptive personalized playlists based on sequence modeling. MIS quarterly, 43(3), 2019

    2019

  69. [77]

    Using retweets when shaping our online persona: Topic modeling approach

    Hilah Geva, Gal Oestreicher-Singer, and Maytal Saar-Tsechansky. Using retweets when shaping our online persona: Topic modeling approach. MIS Quarterly, 43(2):501–524, 2019

    2019

  70. [78]

    Logistic regression

    Todd G Nick and Kathleen M Campbell. Logistic regression. Topics in biostatistics, pages 273–301, 2007

    2007

  71. [79]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pages 785–794, 2016

    2016

  72. [80]

    Racial bias in pain assessment and treatment recommendations, and false beliefs about biological differences between blacks and whites

    Kelly M Ho ffman, Sophie Trawalter, Jordan R Axt, and M Norman Oliver. Racial bias in pain assessment and treatment recommendations, and false beliefs about biological differences between blacks and whites. Proceedings of the National Academy of Sciences, 113(16):4296–4301, 2016

    2016

  73. [81]

    Gender stereotypes and workplace bias

    Madeline E Heilman. Gender stereotypes and workplace bias. Research in organizational Behavior, 32:113–135, 2012

    2012

  74. [82]

    Machine learning and health care disparities in dermatology

    Adewole S Adamson and Avery Smith. Machine learning and health care disparities in dermatology. JAMA dermatology, 154(11):1247–1248, 2018

    2018

  75. [83]

    recall (TPR ˆY|Y′, A) versus precision (PPV ˆY|Y′, A) ratio

    Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. A survey on bias and fairness in machine learning. ACM Computing Surveys (CSUR), 54(6):1–35, 2021. 22 A preprint - December 11, 2024 Appendices Appendix A Theoretical Proofs In this appendix...

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.