CoChat
Sign Up
Cited by 0Ethics and Social Impacts of AIAI and HR TechnologiesExplainable Artificial Intelligence (XAI)

The Impact of Structured Validation and Audit Frameworks on the Fairness and Efficiency of AI-Driven Hiring Systems

Open access
MS Industrial and Organizational Psychology from Missouri State University Department of Psychology and SHRM- CP, USA, Ahmad, Safeer·2025·International Journal of Research and Applied Innovations
DOI 10.15662/ijrai.2025.0806023
Open this paper in CoChat — free accountOpen PDF

Abstract

ABSTRACT: Organizations increasingly use artificial intelligence (AI) to screen and rank job applicants, yet these systems can produce disparate outcomes across protected groups and may add operational friction when they require manual review or compliance documentation. Prior work has proposed algorithmic auditing and assurance frameworks, but empirical evidence linking audit intensity to both fairness and hiring efficiency remains limited within the same operational context. Building on internal algorithmic auditing guidance (Raji et al., 2020) and audit systematization in recruitment (Kazim et al., 2021), this study evaluates whether structured validation and audit frameworks are associated with improved fairness and reduced time to hire. Using an applicant-tracking-system dataset from a logistics supply- chain employer (N = 1,906 active applicants; n = 400 hires), we compared a pre-AI manual baseline (2023) with two 2024 AI screening configurations: a compliance-only audit and an assurance-level audit. Fairness was operationalized with adverse impact ratios (AIRs; UGESP, 1978), and efficiency was operationalized as time to hire (days). Results indicated that assurance-level auditing coincided with substantial improvements in AIRs (e.g., minority AIR increased from 0.12 in the manual baseline to 0.85 under assurance) while also reducing time to hire (M difference = 12.87 days relative to the baseline). Logistic and linear models controlling for job family supported these patterns. Findings suggest that structured, higher-intensity audit frameworks can be associated with simultaneous gains in fairness and process efficiency, warranting replication in multi-site, longitudinal studies under emerging audit mandates (e.g., NYC Local Law 144).

AI snapshot

Written from the paper’s full text.

TL;DR

This quasi-experimental single-organization study compares a 2023 manual hiring baseline with two 2024 AI screening configurations—compliance-only audit and assurance-level audit—and finds that higher audit intensity coincided with improved adverse impact ratios and shorter time to hire. The authors conclude that structured assurance-level auditing may be associated with simultaneous fairness and efficiency gains, while cautioning that causal inference is limited by the observational design.

Key findings

  • Fairness improved monotonically with audit intensity: female-to-male AIR rose from 0.19 in the manual baseline to 0.29 under compliance-only audit and 0.66 under assurance-level audit; protected-minority-to-nonminority AIR rose from 0.12 to 0.45 to 0.85, respectively.
  • Chi-square associations between demographic status and hiring outcomes weakened under assurance: gender association was significant in manual baseline (χ²(1) = 61.86, p < .001, φ = 0.257) and compliance-only (χ²(1) = 24.29, p < .001, φ = 0.224) but not under assurance (χ²(1) = 3.55, p = .059, φ = 0.087); protected-minority association was significant in manual baseline (χ²(1) = 117.05, p < .001, φ = 0.351) and compliance-only (χ²(1) = 15.36, p < .001, φ = 0.178) but not under assurance (χ²(1) = 0.62, p = .431, φ = 0.036).
  • Time to hire differed across conditions: one-way ANOVA F(2, 397) = 58.33, p < .001, η² = 0.227. Compared with manual baseline, compliance-only hires were faster by M difference = 9.71 days, 95% CI [7.21, 12.21], t(249.06) = 7.65, p < .001, d = 0.86; assurance-level hires were faster by M difference = 12.87 days, 95% CI [10.50, 15.24], t(268.36) = 10.70, p < .001, d = 1.16.
  • Logistic models controlling for job family showed a manual-baseline female disadvantage (OR = 0.108, 95% CI [0.061, 0.190], p < .001) reduced under assurance (condition-by-female interaction OR = 5.05, 95% CI [2.24, 11.41], p < .001), and a manual-baseline protected-minority disadvantage (OR = 0.035, 95% CI [0.019, 0.064], p < .001) attenuated under assurance (interaction OR = 20.81, 95% CI [9.23, 46.92], p < .001). Linear regression controlling for job family found compliance-only auditing associated with 9.64 fewer days (p < .001) and assurance-level auditing with 12.86 fewer days (p < .001) relative to manual baseline.

Methods

  • Design: quasi-experimental, nonequivalent-groups comparative design within a single logistics supply-chain organization, comparing three conditions: Manual23 (2023 manual screening), AI24-Comp (2024 AI screening with compliance-only audit), and AI24-Assur (2024 AI screening with assurance-level audit).
  • Data/sample: applicant-tracking-system dataset, described as a de-identified, distribution-preserving replica of the organization’s ATS extract. Final analytic sample for selection outcomes was N = 1,906 active applicants; efficiency analysis included n = 400 hires with observed time-to-hire values. Condition sample sizes were 949 (manual baseline), 484 (AI + compliance-only), and 473 (AI + assurance-level). Applicants were included if they applied to Warehouse, Logistics Coordinator, or Driver job families during study windows with complete demographic and outcome data; withdrawn applicants were excluded from fairness analyses but retained for descriptive attrition reporting.
  • Measures and analysis: fairness operationalized as adverse impact ratios (AIRs; UGESP four-fifths rule threshold AIR < 0.80); efficiency operationalized as time to hire in days from application date to offer acceptance date among hired candidates. Analyses included descriptive statistics, one-way ANOVA and Welch’s t tests for time to hire, chi-square tests for demographic-hiring associations within condition, and logistic/linear regression models controlling for job family.

Limitations

  • The paper states the design is observational and based on a single organization; audit configurations were not randomly assigned, so unobserved differences between requisitions may contribute to observed effects. Although models controlled for job family, other contextual factors such as labor market conditions, recruiter staffing, or requisition urgency were not measured.
  • The paper states demographic analyses were limited to gender and a binary protected-minority indicator; small subgroup sizes (e.g., nonbinary candidates) limited assessment of intersectional effects.
  • The paper states fairness was evaluated using selection outcomes rather than downstream job performance or retention, so it does not address whether audit frameworks improve predictive validity or long-term workforce outcomes; time to hire captures only one dimension of operational efficiency, and additional metrics such as cost per hire, recruiter workload, candidate experience, and quality of hire are needed for a complete cost-benefit assessment.
  • The text also makes evident, in the Method section, threats to causal interpretation including history effects, maturation/implementation effects, selection and job-mix confounding, and instrumentation effects; the authors recommend stratified or covariate-adjusted analyses and caution that causal conclusions should be interpreted cautiously.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.15662/ijrai.2025.0806023

Cites

0 citation errors · 0 not checkable

Statistics

Claims the paper makes

14 of 20 cited sources checked · 7 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Commonly used statistical fairness criteria can be mutually incompatible except under restrictive conditions.

    “commonly used statistical fairness criteria can be mutually incompatible except under restrictive conditions”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Kleinberg et al., 2017 source
  • Premisecitation

    Existing literature often treats fairness and performance as a trade-off problem.

    “Existing literature often treats fairness and performance as a trade-off problem”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Kleinberg et al., 2017 source

      Section 1.2 (p. 6) discusses optimizing the accuracy of risk assignments subject to fairness conditions, so the paper connects fairness with performance. But it identifies its central trade-offs as conflicts between different definitions of fairness, not between fairness and performance, and does not establish how often the broader literature frames the issue that way. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
  • Findingown data

    A one-way ANOVA indicated a significant condition effect on time to hire, F(2, 397) = 58.33, p < .001, η2 = 0.227.

    “A one-way ANOVA indicated a significant condition effect on time to hire, F(2, 397) = 58.33,p < .001,η² = 0.227.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The time-to-hire timestamps were recorded consistently across conditions.
  • Findingown data

    The assurance-level audit significantly reduced the gender disparity, as indicated by a positive condition-by-female interaction (OR = 5.05, 95% CI [2.24, 11.41], p < .001).

    “The assurance-level audit significantly reduced the gender disparity, as indicated by a positive condition-by-female interaction (OR = 5.05, 95% CI [2.24, 11.41], p < .001).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The audit conditions were implemented as specified in the operationalization matrix.
    • Rests on an assumption: The job-family control adequately addresses confounding by job-family composition.
  • Findingown data

    In a linear regression model for time to hire controlling for job family, compliance-only auditing was associated with 9.64 fewer days relative to the manual baseline (p < .001).

    “A linear regression model for time to hire controlling for job family yielded similar conclusions: relative to the manual baseline, compliance-only auditing was associated with 9.64 fewer days (p < .001)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The time-to-hire timestamps were recorded consistently across conditions.
    • Rests on an assumption: The job-family control adequately addresses confounding by job-family composition.
  • Classified as probably not a substantial claim. 23
  • May combine two claims. 16

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Analyses of the early audit regime highlight concerns about definitional scope, methodological variability, and the risk that compliance may prioritize minimum reporting over substantive fairness improvement.

    “analyses of the early audit regime highlight concerns about definitional scope, methodological variability, and the risk that compliance may prioritize minimum reporting over substantive fairness improvement”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Filippi et al., 2023 source

      The Introduction (p. not provided) describes the law’s scope and changes to its proposed audit metrics; §5.2 shows that the prescribed metrics can miss bias. Section 4 says the law mandates computing metrics but specifies no fairness threshold. These passages support concerns about the scope and effectiveness of the audit methodology, but do not establish that compliance prioritizes minimum reporting over substantive fairness improvement. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites Groves et al., 2024 source
    • cites Wright et al., 2024 source

      The Introduction (p. 2) identifies uncertainty about which systems fall within LL 144, and §6.2 (p. 11) finds that published audit reports offered minimal explanation. The Introduction (p. 2) also says the regime makes it impossible to tell whether employment discrimination is declining, supporting the concern that reporting may not yield substantive fairness improvements. The passages reviewed do not establish methodological variability among audits; differences in what reports disclose are not necessarily differences in audit methods. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Kazim et al. (2021) systematized audit stages for algorithmic recruitment, aligning technical checks with organizational governance.

    “Kazim et al. (2021) systematized audit stages for algorithmic recruitment (e.g., purpose definition, data and feature review, validation, fairness testing, and monitoring), aligning technical checks with organizational governance.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Kazim et al., 2021 source

      Section 3.2, “Auditing Stages,” organizes recruitment audits around data, model, and development checks, including feature choices, performance, fairness, and robustness. Section 3 also places technical assessments within a wider audit covering the business case and organizational responsibilities. The cited paper does not set out the sentence’s full example sequence—particularly monitoring—as audit stages. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Systematic auditing approaches document design decisions, risk assumptions, and failure modes across the full system lifecycle.

    “systematic auditing approaches that document design decisions, risk assumptions, and failure modes across the full system lifecycle”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Raji et al., 2020 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Raji et al. (2020) proposed an internal algorithmic auditing framework spanning data collection, model development, testing, deployment, and monitoring, with stage-specific artifacts that enable traceability and accountability.

    “Raji et al. (2020) proposed an internal algorithmic auditing framework spanning data collection, model development, testing, deployment, and monitoring, with stage-specific artifacts that enable traceability and accountability.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Raji et al., 2020 source

      Section 4 presents SMACTR as an internal auditing framework with documentation and artifacts for each stage; Section 4.1 describes a documentation ‘transparency trail’ across development. But its five stages are Scoping, Mapping, Artifact Collection, Testing, and Reflection—not the claimed sequence of data collection through monitoring. The copy discusses deployment and testing of already-launched products, but does not define monitoring as a framework stage. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Recruitment audit frameworks have been proposed and systematized.

    “while recruitment audit frameworks have been proposed and systematized”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Kazim et al., 2021 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    AI Fairness 360 provides bias metrics and mitigation methods across the machine-learning pipeline.

    “Toolkits such as AI Fairness 360 operationalize these concepts by providing bias metrics and mitigation methods across the machine-learning pipeline”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Bellamy et al., 2018 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The audit market is heterogeneous and audit quality depends on clear standards, independence, and transparent reporting practices.

    “Field scans of the auditing ecosystem further note that the audit market is heterogeneous and that audit quality depends on clear standards, independence, and transparent reporting practices”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Costanza-Chock et al., 2022 source

  • May combine two claims.

    Premisecitation

    Early analyses of audit disclosures suggest variability in rigor, method choice, and interpretability, raising questions about whether compliance-only audits are sufficient to meaningfully reduce disparate outcomes in practice.

    “early analyses of audit disclosures suggest variability in rigor, method choice, and interpretability, raising questions about whether compliance-only audits are sufficient to meaningfully reduce disparate outcomes in practice”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Filippi et al., 2023 source

      The Introduction and §5.2 (page numbers unavailable) show that the choice of audit metric affects whether bias is detected, including in real recruitment data. The paper does not analyze audit disclosures or establish variability in auditors’ rigor or interpretability, and it does not test whether compliance-only audits reduce disparate outcomes in practice. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites Groves et al., 2024 source
    • cites Wright et al., 2024 source

      Section 6.2.4 (p. 11) finds that the published audit reports lacked enough explanation for job seekers to interpret them, and the Introduction (p. 2) says the available evidence cannot establish whether LL 144 reduces discrimination. These passages support the concern about interpretability and compliance-only audits, but the paper does not establish variability in audit rigor or method choice across disclosures. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Condition sample sizes were 949 for manual baseline, 484 for AI + compliance-only, and 473 for AI + assurance-level.

    “Condition sample sizes were 949 (manual baseline), 484 (AI + compliance-only), and 473 (AI + assurance-level).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The audit conditions were implemented as specified in the operationalization matrix.

  • Classified as probably not a substantial claim.

    Findingown data

    The efficiency analysis included 400 hires with observed time-to-hire values.

    “the efficiency analysis included n = 400 hires with observed time-to-hire values”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.

  • May combine two claims.

    Findingown data

    The compliance-only audit produced meaningful gains relative to the manual baseline, and the assurance-level audit produced the largest reductions in disparity.

    “the compliance-only audit produced meaningful gains relative to the manual baseline, and the assurance-level audit produced the largest reductions in disparity.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The audit conditions were implemented as specified in the operationalization matrix.

  • May combine two claims.

    Findingown data

    For protected-minority status, the minority-to-nonminority AIR increased from 0.12 in the manual baseline to 0.45 under the compliance-only audit and to 0.85 under the assurance-level audit.

    “For protected-minority status, the minority-to-nonminority AIR increased from 0.12 in the manual baseline to 0.45 under the compliance-only audit and to 0.85 under the assurance-level audit (Table 5; Figure 4B).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The audit conditions were implemented as specified in the operationalization matrix.

  • May combine two claims.

    Findingown data

    The minority disadvantage was substantially attenuated under both audits, with the largest reduction under assurance (interaction OR = 20.81, 95% CI [9.23, 46.92], p < .001).

    “was substantially attenuated under both audits, with the largest reduction under assurance (interaction OR = 20.81,95% CI [9.23, 46.92], p < .001).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The audit conditions were implemented as specified in the operationalization matrix.
    • Rests on an assumption: The job-family control adequately addresses confounding by job-family composition.

  • Classified as probably not a substantial claim.

    Findingown data

    The final analytic sample for selection outcomes included 1,906 active applicants.

    “The final analytic sample for selection outcomes included N = 1,906 active applicants”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.

  • May combine two claims.

    Findingown data

    Time to hire decreased substantially under both audit conditions, with the largest reduction under assurance.

    “At the same time, time to hire decreased substantially under both audit conditions, with the largest reduction under assurance.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.
    • Rests on an assumption: The time-to-hire timestamps were recorded consistently across conditions.

  • May combine two claims.

    Study design and conductnone given

    The design is observational and based on a single organization; audit configurations were not randomly assigned.

    “First, the design is observational and based on a single organization; audit configurations were not randomly assigned”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Using AIRs provides an interpretable bridge between statistical fairness evaluation and employment-law practice.

    “From a measurement perspective, using AIRs provides an interpretable bridge between statistical fairness evaluation and employment-law practice (UGESP, 1978).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites UGESP, 1978

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationown data

    Observed efficiency gains suggest assurance-level audits need not be interpreted solely as compliance overhead; they may function as process-improvement mechanisms that reduce cycle time.

    “Importantly, the observed efficiency gains suggest that assurance-level audits need not be interpreted solely as compliance overhead; they may function as process-improvement mechanisms that reduce cycle time.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation

    Assurance-level auditing likely involved more than reporting selection-rate ratios.

    “In practical terms, assurance-level auditing likely involved more than reporting selection-rate ratios.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Governance interventions can target process inefficiencies while also improving equity.

    “suggesting that governance interventions can target process inefficiencies (e.g., rework, escalations, and uncertainty about model use) while also improving equity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    In this dataset, higher audit intensity coincided with both improved fairness and faster hiring.

    “In this dataset, higher audit intensity coincided with both improved fairness and faster hiring”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    Such activities can plausibly improve both fairness and efficiency by reducing false negatives for qualified candidates in protected groups and by stabilizing decision thresholds that otherwise require manual override.

    “Such activities can plausibly improve both fairness and efficiency by reducing false negatives for qualified candidates in protected groups and by stabilizing decision thresholds that otherwise require manual override.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Models controlled for job family.

    “Although models controlled for job family”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    AIRs do not identify which stage(s) of the hiring pipeline generate disparities, nor do they address other fairness constructs such as calibration or equal opportunity.

    “However, AIRs do not identify which stage(s) of the hiring pipeline generate disparities, nor do they address other fairness constructs (e.g., calibration or equal opportunity).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Unobserved differences between requisitions may contribute to observed effects.

    “so unobserved differences between requisitions may contribute to observed effects.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    New York City's Local Law 144 requires annual independent audits and candidate notifications for covered AEDTs.

    “New York City's Local Law 144 requires annual independent audits and candidate notifications for covered AEDTs”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Trustworthy AI requires proactive risk assessment, monitoring, and accountability mechanisms rather than ad hoc post hoc checks.

    “trustworthy AI requires proactive risk assessment, monitoring, and accountability mechanisms rather than ad hoc post hoc checks.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites NIST, 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites ISO/IEC 42001:2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisenone given

    New York City's Local Law 144 has mandated independent bias audits for certain automated employment decision tools (AEDTs).

    “regulatory efforts such as New York City's Local Law 144 have mandated independent bias audits for certain automated employment decision tools (AEDTs)”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    These critiques suggest that compliance-only audits may not adequately address deeper sources of bias (e.g., construct validity, proxy features, and organizational process effects).

    “These critiques suggest that compliance-only audits may not adequately address deeper sources of bias (e.g., construct validity, proxy features, and organizational process effects)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Wilson et al. (2021) described a candidate-screening system that incorporated fairness evaluation and external review as part of product development.

    “Wilson et al. (2021) described a candidate-screening system that incorporated fairness evaluation and external review as part of product development.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Wilson et al., 2021 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The study does not address whether audit frameworks improve predictive validity or long-term workforce outcomes.

    “the study does not address whether audit frameworks improve predictive validity or long-term workforce outcomes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Time to hire captures only one dimension of operational efficiency.

    “Fourth, time to hire captures only one dimension of operational efficiency”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Fairness was evaluated using selection outcomes rather than downstream job performance or retention.

    “fairness was evaluated using selection outcomes rather than downstream job performance or retention”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The Uniform Guidelines on Employee Selection Procedures (UGESP, 1978) describe how employers should evaluate whether selection procedures create adverse impact, including the four-fifths rule as a practical screening criterion.

    “The Uniform Guidelines on Employee Selection Procedures (UGESP, 1978) describe how employers should evaluate whether selection procedures create adverse impact, including the four-fifths rule as a practical screening criterion.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites UGESP, 1978

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

34 of 34

  • The paper relies on this without support.

    The de-identified analytic dataset is distribution-preserving relative to the organization's original ATS extract.

    “the analytic dataset is a de-identified, distribution-preserving replica of the organization's ATS extract.”

  • The paper relies on this without support.

    The audit conditions were implemented as specified in the operationalization matrix.

    “This matrix defines, at an implementation level, what compliance-only auditing (AI24-Comp) versus assurance-level auditing (AI24-Assur) meant in this study.”

  • The paper relies on this without support.

    The time-to-hire timestamps were recorded consistently across conditions.

    “All timestamps should be recorded consistently across conditions.”

  • The paper relies on this without support.

    The job-family control adequately addresses confounding by job-family composition.

    “to reduce confounding from job-family composition differences across conditions, generalized linear models (logistic regression) predicted hiring outcomes from condition, demographic status, and their interaction while controlling for job family”
4 of 4

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.

  • ConsistentF(2, 397) = 58.33reported p < .001 · recomputed p < .001

    “F(2, 397) = 58.33,p < .001”

  • Consistentt(249.06) = 7.65reported p < .001 · recomputed p < .001

    “t(249.06)= 7.65, p < .001”

  • Consistentt(268.36) = 10.7reported p < .001 · recomputed p < .001

    “t(268.36)= 10.70,p < .001”

  • Consistentχ²(1) = 61.86reported p < .001 · recomputed p < .001

    “χ2(1) = 61.86,p < .001”

  • Consistentχ²(1) = 24.29reported p < .001 · recomputed p < .001

    “χ2(1) = 24.29,p < .001”

  • Consistentχ²(1) = 3.55reported p = .059 · recomputed p = .060

    “χ2(1) = 3.55, p = .059”

  • Consistentχ²(1) = 117.05reported p < .001 · recomputed p < .001

    “χ2(1) = 117.05,p < .001”

  • Consistentχ²(1) = 15.36reported p < .001 · recomputed p < .001

    “χ2(1) = 15.36,p < .001”

  • Consistentχ²(1) = 0.62reported p = .431 · recomputed p = .431

    “χ2(1) = 0.62, p = .431”

9 of 9

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Self-check results appear here once an independent check has confirmed them.

Cites

The publisher hasn’t deposited this paper’s reference list.

Related

Authors
MS Industrial and Organizational Psychology from Missouri State University Department of Psychology and SHRM- CP, U., & Ahmad, S.
Affiliations
Missouri State University (US)
Corresponding author
Ahmad, Safeer
Venue
International Journal of Research and Applied Innovations
Publisher
Ess & Ess Research Publications
Published
2025
ISSN
2455-1864 (ISSN-L)
DOI
10.15662/ijrai.2025.0806023
OpenAlex
W7123558448
Semantic Scholar
df882dde9acb227149e05908d5736b2a717a6212
Cited by
0
References
0
FWCI
0.00
Type
journal article
Language
en
Open access
Locations
International Journal of Research and Applied Innovations (publishedVersion)
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Business, Management and Accounting › Organizational Behavior and Human Resource Management › AI and HR Technologies
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
Keywords
algorithmic auditing, AI-driven hiring, applicant tracking systems, algorithmic fairness, time-to-hire, recruitment efficiency, compliance audit, protected groups, algorithmic accountability

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-30.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-30.

Cited sources are checked by 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