The DOI matches its Crossref record; 72 of the 73 references it cites were found in a registry or an index; 14 of the 42 sources its claims cite were checked for support.
Human–AI Interactions in Public Sector Decision Making: “Automation Bias” and “Selective Adherence” to Algorithmic Advice
Open accessAbstract
Abstract Artificial intelligence algorithms are increasingly adopted as decisional aides by public bodies, with the promise of overcoming biases of human decision-makers. At the same time, they may introduce new biases in the human–algorithm interaction. Drawing on psychology and public administration literatures, we investigate two key biases: overreliance on algorithmic advice even in the face of “warning signals” from other sources (automation bias), and selective adoption of algorithmic advice when this corresponds to stereotypes (selective adherence). We assess these via three experimental studies conducted in the Netherlands: In study 1 (N = 605), we test automation bias by exploring participants’ adherence to an algorithmic prediction compared to an equivalent human-expert prediction. We do not find evidence for automation bias. In study 2 (N = 904), we replicate these findings, and also test selective adherence. We find a stronger propensity for adherence when the advice is aligned with group stereotypes, with no significant differences between algorithmic and human-expert advice. In study 3 (N = 1,345), we replicate our design with a sample of civil servants. This study was conducted shortly after a major scandal involving public authorities’ reliance on an algorithm with discriminatory outcomes (the “childcare benefits scandal”). The scandal is itself illustrative of our theory and patterns diagnosed empirically in our experiment, yet in our study 3, while supporting our prior findings as to automation bias, we do not find patterns of selective adherence. We suggest this is driven by bureaucrats’ enhanced awareness of discrimination and algorithmic biases in the aftermath of the scandal. We discuss the implications of our findings for public sector decision making in the age of automation. Overall, our study speaks to potential negative effects of automation of the administrative state for already vulnerable and disadvantaged citizens.
Overview
CoChat Summary
TL;DR
Three Dutch survey experiments found no general automation bias: participants were not more likely to follow algorithmic advice than equivalent human-expert advice, and in study 3 they were less likely to follow it after the childcare benefits scandal. Selective adherence was found in study 2—advice was followed more when it matched negative stereotypes about a Moroccan-Dutch teacher—but not in study 3 with civil servants.
Key findings
- Automation bias was not supported across studies 1 and 2: algorithmic versus human-expert advice differences were small and insignificant (study 1 OR = 0.96, p = .876; study 2 OR = 1.16, p = .498; pooled N = 1,509 OR = 1.07, p = .683).
- In study 3 (N = 1,345 civil servants), participants were significantly less likely to follow algorithmic advice than human advice (OR = 0.54, p = .006; 4.8% vs. 8.6% selected the teacher with the lowest ILE score).
- Study 2 (N = 792 Dutch-descent respondents) found selective adherence: a Moroccan-Dutch teacher with a low ILE score was 50% more likely not to have their contract renewed than a Dutch teacher with the same score (OR = 1.50, p = .04 one-sided; 12.3% vs. 8.6%).
- The selective-adherence effect was not exacerbated by algorithmic versus human advice (interaction OR = 0.50, |z| = 1.45, p = .147), and study 3 did not replicate selective adherence: the Moroccan-Dutch teacher was less likely to be sanctioned (OR = 0.57, p = .020).
Methods
- Three preregistered survey experiments in the Netherlands: study 1 (N = 605) and study 2 (N = 904) among Dutch citizens; study 3 (N = 1,345) among Dutch civil servants.
- Participants acted as school board members deciding which of three teachers not to renew, using a qualitative HR evaluation and a numeric ILE score; the score was described as produced by an algorithm or human experts, and studies 2–3 manipulated the low-scoring teacher’s name as Dutch or Moroccan-Dutch.
- Main analysis used logistic regression on a binary outcome of following the ILE score; two-tailed p values are reported uniformly, with one-sided p values for preregistered directional hypotheses.
Limitations
- The paper states study 3’s sample was not fully representative of the Dutch public sector (women underrepresented, higher mean age), lacked an attention check, and may be affected by social desirability bias; it also notes statistical power limitations for the H3 interaction analysis.
- The paper states study 3 coincided with the childcare benefits scandal, so results may reflect heightened awareness of algorithmic bias and discrimination rather than stable effects.
- Evident from the text: all studies use a hypothetical vignette about teacher employment, which may limit external validity to other public-sector decisions and real-world settings.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
9 Valid · 4 Not valid · 21 Not checked · 135 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Additionally, the study finds that judges were more likely to accept low scores for white defendants, while overriding similar scores for black defendants.”
What the sources say
- Disputedcites Stevenson (2018) sourceExperimental — accuracy not yet measured
The Introduction says white defendants appeared to benefit more, but attributes that difference to regional variation in judges’ responses, not to judges accepting low scores for white defendants and overriding comparable scores for black defendants. It says white and black defendants within the same county saw similar increases in release. Section C discusses differential use of scores as a possibility, not a finding of the study. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
hedgedstates an association- 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.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Following a motivated reasoning logic, this growing body of literature has established that decision-makers are prone to selectively seek and interpret information in light of preexisting stereotypes, beliefs, and social identities.”
What the sources say
- Partly supportedcites Baekgaard et al. 2019 sourceExperimental — accuracy not yet measured
The Discussion (page not provided) says politicians’ prior attitudes affect how they interpret information, and Experiments 1 and 2 report poorer interpretations when information conflicts with those attitudes. The Discussion describes selective attention under information overload as a possible explanation, but the paper does not establish that decision-makers selectively seek information or that social identities drive the effect. (LLM, conf=0.85)
- Could not verifycites Baekgaard and Serritzlew 2016 sourceExperimental — accuracy not yet measured
The available passages contain bibliographic metadata but no discussion of selective information seeking or interpretation. Searches for motivated reasoning, preexisting beliefs, stereotypes, and social identities returned no substantive passage from the paper, so our copy cannot verify the claim. (LLM, conf=0.76)
- Could not verifycites Christensen et al. 2018 sourceExperimental — accuracy not yet measured
Searches for motivated reasoning, selective interpretation, stereotypes, and social identities found no relevant passage from the paper’s full text. The indexed results returned only abstract or bibliographic material, which cannot establish this verdict under the specified evidence rule. (LLM, conf=0.69)
boostedstates 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.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Experimental lab studies have diagnosed this tendency across a number of research fields (Goddard, Roudsari, and Wyatt 2012).”
What the sources say
- Supportedcites Goddard, Roudsari, and Wyatt 2012 source
The Abstract says automation bias has been studied in various academic fields, and the research-field table lists studies in healthcare, aviation, human–computer interaction, and other fields. The Confidence and trust section discusses experimental studies that found automation-bias errors, supporting the sentence’s broad characterization. (LLM, conf=0.84)
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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“We specifically selected this ethnic minority group in the Netherlands, since it is a minority group that is often negatively stereotyped (Jilke, Van Dooren, and Rys 2018; Kamans et al. 2009).”
What the sources say
- Not matchedcites Jilke, Van Dooren, and Rys 2018
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites Kamans et al. 2009 source
The META-STEREOTYPES AND THEIR INFLUENCE ON BEHAVIOR section says that people of Moroccan descent in Dutch society are confronted with a clearly negative stereotype. The paper also describes Dutch Moroccan teenagers as facing negative stereotypes about their group, supporting the citing sentence’s background claim. (LLM, conf=0.94)
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.
- Not matched
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These findings are also in line with results from other studies on pretrial algorithmic risk scores by law and computer science scholars respectively, which report patterns consistent with biased adherence to algorithmic advice (Green and Chen 2019a, 2019b; Stevenson 2018).”
What the sources say
- Not matchedcites Green and Chen 2019a, 2019b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites Stevenson 2018 source
The Introduction reports that Kentucky judges ignored the risk tool’s presumptive default of non-monetary release in more than two-thirds of cases, while changing release practices differently across risk groups. The section ‘Lessons for Other Jurisdictions’ likewise says judges deviated from the tool’s action-directives more often than not. These findings support the hedged description of patterns consistent with biased adherence, though the paper does not establish judges’ motives. (LLM, conf=0.86)
plainstates an association- 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.
- Not matched
82 more claims · 1 with their cited sources checked
Create a free account to see all 169 claims and to verify the 5 references still open.
- Probably not a substantial claim. 42
- It may state two separate claims as one. 48
- The quote only partly states this claim. 5
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
“Even when AI operates at the level of recommendation rather than direct action, the mere presence of an algorithmic suggestion can systematically shift human judgment toward the algorithm's conclusion, a phenomenon known as automation bias (Goddard, Roudsari, and Wyatt 2012;Alon-Barkat and Busuioc 2023).”
From Limits of Interventions in AI-Assisted Decision Making (2026) · cites this paper as Alon-Barkat and Busuioc 2023
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“In contrast to all these findings, Alon-Barkat and Busuioc (2023) found selective adherence to AI suggestions, rather than AB, in employment decisions.”
From Exploring automation bias in human–AI collaboration: a review and implications for explainable AI (2026) · cites this paper as Alon-Barkat and Busuioc 2023
The cited source says what the claim attributes to it.
In the citing paper
“In this study, participants were more likely to follow advice aligned with their pre-existing stereotypes (e.g., against ethnic minority indicators like a Moroccan-sounding name): they were more inclined to terminate contracts of Moroccan-sounding teachers when scores were low, despite similar scores for Dutch teachers.”
From Exploring automation bias in human–AI collaboration: a review and implications for explainable AI (2026) · cites this paper as Alon-Barkat and Busuioc 2023
The cited source says what the claim attributes to it.
In the citing paper
“Our research has also led to a subset of studies (Alon-Barkat and Busuioc 2023; Selten etal. 2023; Bashkirova and Krpan 2024) that challenge the prevailing belief of AB as a distinct phenomenon in human-AI interaction.”
From Exploring automation bias in human–AI collaboration: a review and implications for explainable AI (2026) · cites this paper as Alon-Barkat and Busuioc 2023
The cited source says what the claim attributes to it.
In the citing paper
“This finding aligns with theoretical debates on automation and discretion, where concerns about 'black box' systems and loss of human oversight have been central (Alon- Barkat and Busuioc 2023).”
From Algorithmic Governance: Experimental Evidence on Citizens' and Public Administrators' Legitimacy Perceptions of Automated Decision‐Making (2026) · cites this paper as Alon- Barkat and Busuioc 2023
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- Consistentχ²(1) = 2.141reported p = .143 · recomputed p = .143
“χ 2 (1) = 2.141, p = .143”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Selective Adherence · p. 9
“Comparing the two models via BIC and AIC indicates that the main effect model is more appropriate (Lorah 2020).”
§ Study 3: Replication with a Sample of Civil Servants · p. 10
“| AIC | 529.579 | | | 529.438 | | |”
Citations
Cites
Rudin, C. (2019) · Nature Machine Intelligence · doi:10.1038/s42256-019-0048-x · Cited by 10,561
Humans and automation: Use, misuse, disuse, abuse
Parasuraman, R., & Riley, V. (1997) · Human Factors: The Journal of the Human Factors and Ergonomics Society · doi:10.1518/001872097778543886 · Cited by 4,554 · cited 1×
“…automated systems by human actors that disregard contradictory information from other sources or do not (thoroughly) search for additional information (Cummings 2006; Lyell and Coiera 2017; Mosier et al. 2001; Parasuraman and Riley 1997; Skitka, Mosier, and Burdick 1999, 2000; Skitka et al. 2000).”
Background(2018) · Cited by 3,510
Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor
Lebovits, H. (2019) · Public Integrity · doi:10.1080/10999922.2018.1511671 · Cited by 3,279 · cited 5×
“Such concerns become particularly relevant given welldocumented failures and malfunctioning of AI(-informed) systems (e.g., Benjamin 2019; Buolamwini and Gebru 2018; Ferguson 2017; Eubanks 2018; O’Neill 2016; Richardson, Schultz, and Crawford 2019; Rudin 2019).”
BackgroundInfluentialDoyle, T. (2017) · The Information Society · doi:10.1080/01972243.2017.1354593 · Cited by 3,071
Algorithm appreciation: People prefer algorithmic to human judgment
Logg, J. M., Minson, J. A., & Moore, D. A. (2019) · Organizational Behavior and Human Decision Processes · doi:10.1016/j.obhdp.2018.12.005 · Cited by 2,061
Review of “Race After Technology: Abolitionist Tools for the New Jim Code”
Jipguep-Akhtar, M. (2020) · Social Forces · doi:10.1093/sf/soz162 · Cited by 1,473
Automation bias: a systematic review of frequency, effect mediators, and mitigators
Goddard, K., Roudsari, A., & Wyatt, J. C. (2012) · Journal of the American Medical Informatics Association · doi:10.1136/amiajnl-2011-000089 · Cited by 1,162
Bovens, M., & Zouridis, S. (2002) · Public Administration Review · doi:10.1111/0033-3352.00168 · Cited by 964
Algorithmic regulation: A critical interrogation
Yeung, K. (2018) · Regulation & Governance · doi:10.1111/rego.12158 · Cited by 842
Research works printed but not linked: 9
- Angwin, 2016, Machine bias., ProPublica
- Geiger, 2021, How a discriminatory algorithm wrongly accused thousands of families of fraud., Vice
- Israni, 2017, When an algorithm helps send you to prison., New York Times
- Medium—Open Letter Concerned AI Researchers, 2019, On recent research auditing commercial facial analysis technology, Medium
- Milner, 2016, Death by GPS: Are Satnavs changing our brains?, The Guardian
- O’Neil, 2016, Weapons of math destruction: How big data increases inequality and threatens democracy
- Richardson, 2019, Dirty data, bad predictions: How civil rights violations impact police data, predictive policing systems, and justice., New York University Law Review Online, 94, 192
- Turque, 2012, ‘Creative...motivating’ and fired., Washington Post
- Volkskrant, 2020, Belastingdienst schuldig aan structurele discriminatie van mensen die toeslagen ontvingen, Volkskrant
Cited by
Exploring artificial intelligence adoption in public organizations: a comparative case study
Neumann, O., Guirguis, K., & Steiner, R. (2024) · Public Management Review · doi:10.1080/14719037.2022.2048685 · Cited by 451
“Lastly, Eubanks (2017) as well as Alon-Barkat and Busuioc (2022) explored how automated decision-making in public services may negatively impact already disad-vantaged groups and reinforce existing biases, while Bannister and Connolly (2020) provided a taxonomy of decision-making algorithms in…”
BackgroundExploring automation bias in human–AI collaboration: a review and implications for explainable AI
Romeo, G., & Conti, D. (2026) · AI & SOCIETY · doi:10.1007/s00146-025-02422-7 · Cited by 209
Gender bias perpetuation and mitigation in AI technologies: challenges and opportunities
O’Connor, S., & Liu, H. (2024) · AI & SOCIETY · doi:10.1007/s00146-023-01675-4 · Cited by 165
AI Fairness in Data Management and Analytics: A Review on Challenges, Methodologies and Applications
Chen, P., Wu, L., & Wang, L. (2023) · Applied Sciences · doi:10.3390/app131810258 · Cited by 155
Selten, F., Robeer, M., & Grimmelikhuijsen, S. (2023) · Public Administration Review · doi:10.1111/puar.13602 · Cited by 153
Metadata
- Authors
- Alon-Barkat, S., & Busuioc, M.
- Affiliations
- University of Haifa , Israel
- Vrije Universiteit Amsterdam (NL)
- Corresponding author
- Busuioc, Madalina
- Venue
- Journal of Public Administration Research and Theory
- Publisher
- Oxford University Press (OUP)
- Published
- 2022-02-08
- Volume
- 33
- Issue
- 1
- Pages
- 153-169
- ISSN
- 1053-1858 (ISSN-L)
- 1477-9803
- OpenAlex
- W4211002924
- Cited by
- 469
- References
- 73
- Citations by year
- 2026: 235 · 2025: 120 · 2024: 68 · 2023: 35 · 2022: 11
- FWCI
- 67.76
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Journal of Public Administration Research and Theory (publishedVersion, cc-by-nc)
- VU Research Portal (publishedVersion, other-oa)
- arXiv (Cornell University) (submittedVersion)
- Data Archiving and Networked Services (DANS) (submittedVersion)
- arXiv (Cornell University)
- Funders
- European Union’s Horizon 2020 (716439)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Political Science and International Relations › E-Government and Public Services
- Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
- Keywords
- automation bias, algorithmic decision-making, human-AI interaction, public sector decision-making, algorithmic advice, group stereotypes, public sector employees, algorithmic discrimination, administrative automation
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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (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: Human–AI Interactions in Public Sector Decision Making: “Automation Bias” and “Selective Adherence” to Algorithmic Advice. https://app.cochat.ai/papers/10.1093/jopart/muac007
- The paper
- Alon-Barkat, S., & Busuioc, M. (2023). Human–AI Interactions in Public Sector Decision Making: “Automation Bias” and “Selective Adherence” to Algorithmic Advice. Journal of Public Administration Research and Theory. https://doi.org/10.1093/jopart/muac007
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.