Talent Systems — Science Team
Phases Awaiting Your Input

Phase 14: Adverse Impact Methodology

What needs to be designed for EEOC/OFCCP adverse impact reporting.

Time estimate: ~1-2 weeks
Impact: Compliance reporting
Dependencies: Phase 13 (validity data)

The current per-candidate adverse impact check in the scorer prompt is NOT real adverse impact analysis. Real adverse impact is aggregate across subgroups using the 4/5ths rule. This phase replaces the per-candidate check with proper methodology.

What We Need

1. Demographic Categories

Specify EEOC standard categories for self-reporting:

  • Race/ethnicity groupings
  • Gender categories
  • Age ranges
  • Disability status
  • Veteran status
  • Any additional categories

2. Statistical Tests

Specify which tests to implement:

  • Standardized mean differences (Cohen's d)
  • Confidence intervals
  • Effect sizes
  • 4/5ths rule (selection rate comparison)
  • What else?

3. Minimum Sample Sizes

Define minimum N thresholds:

  • Per subgroup for meaningful analysis
  • Per posting for reliable results
  • Platform-wide for aggregate reporting

4. Self-Report Collection

Design the candidate self-report flow:

  • When should it be presented? (After interview? Separate link?)
  • How to maximize participation without coercion?
  • Explicit opt-out must be available

5. DIF Analysis

Differential item functioning on questions:

  • Which method? (Mantel-Haenszel, logistic regression, IRT-based?)
  • What constitutes a flagged item?
  • Per-question or per-competency?

6. Remediation Guidance

If adverse impact is detected:

  • What should the platform recommend?
  • Content audit? Question review? Anchor revision?
  • At what severity level?

7. OFCCP-Specific Requirements

Any requirements beyond standard EEOC that OFCCP compliance demands:

  • Documentation format
  • Reporting cadence
  • Record retention

On this page