{"resourceId":"mcconvey-fairness-theatre-ews-20260929","versions":[{"version":"external-cdba245d4f9edcf9915bbc09f2db17a08b2a1ffe3df553ced74128f9340afcf3","resource":{"id":"mcconvey-fairness-theatre-ews-20260929","title":"Fairness Theatre: post-hoc fairness on vendor-controlled college EWS redistributes disparities without fixing them","organization":"University of Toronto & Georgia Institute of Technology (McConvey, Zhai, Li, Guha)","sector":"Higher education advising / early warning systems / algorithmic fairness","geography":"Canada (public college in Ontario; research evaluation under simulated procurement constraints)","publishedAt":"Submitted to arXiv 29 September 2026 (cs.HC)","publicationDate":"2026-09-29","eventDate":null,"sourceName":"arXiv","sourceLabel":"Academic preprint (cs.HC)","sourceUrl":"https://arxiv.org/pdf/2609.38552","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["governance-procurement","data-security","accessibility-workforce","operating-model"],"finding":"Using records from a public college in Ontario, the authors evaluate six post-hoc fairness interventions on a research replica of a proprietary Early Warning System under simulated procurement constraints that leave institutions able to adjust outputs but not model design. Interventions redistributed disparities without consistently reducing them; some approaches favored already-advantaged groups when disadvantage was defined via group size; small groups remained poorly served. They term the resulting condition fairness theatre: institutions perform fairness work while groups’ error burdens persist or worsen because vendors retain design control.","sledRelevance":"Direct HE advising/retention systems signal: equity overlays on vendor-locked early-warning products can look like fairness work without improving who is wrongly flagged or missed. Complements tutoring equity findings by showing procurement structure—not only model choice—limits student-success fairness.","evidence":"Empirical evaluation of six post-hoc methods on college EWS replica with error-type profiling. Simulated constraints; single college; preprint; fairness metrics do not establish retention gains.","architectureImplications":"Prefer systems where institutions can inspect features, labels, and training objectives—not only threshold/post-processing knobs. Log false-positive/negative burdens by group before and after any fairness overlay.","governanceImplications":"Procurement should require model-design transparency and contestability, not only post-hoc fairness dashboards. Treat fairness theatre as a contract failure mode when vendors lock design.","securityPrivacyImplications":"EWS features may include sensitive proxies; minimize, document, and avoid stigmatizing outreach from opaque scores.","caveats":"Single Ontario college; simulated procurement; research replica not live vendor A/B; fairness metrics do not establish that interventions improve retention; preprint.","streamIds":["student-success"],"roles":{"sales":"For student-success and IR buyers evaluating early-warning vendors, Fairness Theatre warns that post-hoc fairness knobs under procurement lock-in can redistribute who is mis-flagged without reducing disparity—and may even favor already-advantaged groups when size defines disadvantage. Sell a procurement and validation engagement that demands design transparency and group-wise error profiles—not a fairness dashboard add-on. Bound claims to this Ontario college evaluation and refuse guaranteed equity from post-processing alone.","engineering":"Engineering should expose feature sets, label definitions, and thresholding, and compute false-positive/negative burdens by group before and after any fairness overlay. Avoid shipping post-hoc-only fairness as the equity story when the model is vendor-locked. Constraints include proprietary EWS opacity and single-college evidence needing local validation. Proof-of-value is reduced harmful error burdens for underserved groups with advisor-usable overrides—not a prettier disparity chart. Protect sensitive proxy features under privacy review.","delivery":"Delivery should inventory EWS error types by demographic group, pilot any fairness overlay with before/after error profiles, and keep advisors empowered to override. Ownership sits with student-success analytics and procurement jointly. Train advisors that a fair score can still miss small groups. Acceptance means published group-wise error reports and contract language for design access or exit. Risks include fairness theatre and stigmatizing outreach from opaque flags."},"roleDetails":{"sales":{"customerProblem":"Colleges buy vendor EWS tools and try to fix inequity with post-hoc fairness settings they do not control at the model layer.","stakeholders":["student success","institutional research","advising leadership","procurement","equity/compliance"],"discoveryQuestions":["Can we change features, labels, and objectives—or only thresholds?","What are false-positive/negative rates by group today?","What happens to small groups under each fairness method?"],"valueHypothesis":"Procurement that requires design transparency and group-wise error accountability prevents fairness theatre from substituting for equitable advising support.","potentialEngagement":"EWS fairness validation and contract redesign using local records and error-type profiling.","unsupportedClaims":["Post-hoc fairness guarantees equity","Risk scores equal need for help","Retention lift from fairness overlays alone"]},"engineering":{"fit":"Vendor or in-house early-warning systems used for college advising outreach under procurement constraints.","architectureIntegration":"Score service → group-wise error monitor → optional post-hoc adjuster with audit log → advisor UI with override; prefer redesign access over post-processing-only.","prerequisites":["historical outcomes labels","group attributes under privacy policy","advisor override channel"],"constraints":["vendor-locked model design","small-group instability","single-site evidence"],"security":"Sensitive demographic/proxy features; minimize; audit who sees risk scores; no automated high-stakes penalties from opaque scores.","proofOfValue":["Reduced harmful FN/FP for underserved groups","Advisor override usability","Documented ability to change model design or exit"]},"delivery":{"work":"Profile baseline errors, test fairness overlays offline, brief advising leads, update procurement language, monitor live error burdens.","dependencies":["IR/data warehouse","vendor cooperation","advising managers","legal/procurement"],"ownership":"Student-success analytics lead with procurement co-ownership of vendor terms.","skillsAdoption":"Advisors interpret group error reports and override; IR maintains fairness monitoring.","governanceCheckpoints":["pre-overlay error baseline","procurement design-access review","quarterly disparity/error audit"],"acceptanceCriteria":["Group-wise FN/FP published internally","Override path live","Contract addresses design opacity"],"risks":["Fairness theatre","Small-group neglect","Stigmatizing flags"]}},"retrievedAt":"2026-10-02T03:21:55Z","enrichedAt":"2026-10-02T03:28:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"When small or disadvantaged groups stay poorly served after fair overlays, advisors inherit unequal error burdens and students face unequal stigma or support. Advising staff need error-profile literacy, not only risk scores.","procurementImplications":"Write contracts requiring feature/label/objective transparency, group-wise error reporting, and institutional rights to change or exit when post-hoc fairness only redistributes harm.","operatingModelImplications":"Pair any EWS go-live with group-wise false-positive/negative reviews and human override paths; do not equate fairness-dashboard activity with equity outcomes.","sourceVerification":{"openedUrl":"https://arxiv.org/pdf/2609.38552","referenceExcerpt":"Interventions redistributed disparities without consistently reducing them","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}