Where Should Hiring Teams Measure Adverse Impact in the Funnel?

Research question

Should recruiting teams measure selection-rate differences only at the final hiring decision, or at every stage of the funnel, and how should they avoid turning a screening ratio into an unsupported conclusion?

This article reviews the federal Uniform Guidelines and agency explanations, then translates their measurement logic into recruiting data controls. It does not decide whether any employer's practice is lawful. The Uniform Guidelines in 29 CFR Part 1607 and the EEOC's detailed questions and answers are the main authorities considered.

Start with the complete selection process

The EEOC questions and answers define the total selection process as the combined effect of procedures leading to an employment decision. The agency explains that adverse impact is determined first for the overall process for each job. When the overall process shows adverse impact, the effects of individual procedures should then be analyzed.

That sequence prevents a common analytical mistake: treating every difference at an intermediate stage as if it were a final finding. Stage data remain useful. They can show where applicants leave the process or where a requirement changes selection rates. Their role is diagnostic, and their interpretation depends on the complete process, the job, the period, and the governing rules.

A funnel should therefore retain both views. The overall view follows a defined applicant cohort through the final decision. The stage view uses the eligible population entering each specific procedure as its denominator. Mixing those denominators can create ratios that look precise but answer no coherent question.

Calculate the signal without overstating it

The federal questions and answers describe four steps: calculate each group's selection rate, identify the highest rate, divide each other rate by that rate, and observe whether a ratio is substantially lower. The often-cited four-fifths threshold is a rule of thumb used by enforcement agencies. The source expressly says it is not a legal definition and does not resolve whether discrimination occurred.

Suppose 120 candidates in one comparison group enter an assessment and 36 advance. Its selection rate is 30 percent. If 80 candidates in another group enter and 32 advance, that group's rate is 40 percent. Dividing 30 by 40 gives an impact ratio of 0.75. That is below four-fifths. The calculation flags the result for review; it does not explain its cause.

Why disposition data come first

An applicant tracking system may record rejected, withdrawn, unresponsive, duplicate, did not attend, and position canceled. Those labels are not interchangeable. A team that changes the meaning of "withdrawn" midway through a campaign also changes its denominator and outcomes.

Before computing selection rates, recruiting operations should document stage entry, pass, fail, withdrawal, administrative closure, and missing status. Each event needs a timestamp and an owner. Reopened applications and candidates considered for multiple requisitions need explicit handling rules. Otherwise, the same person can appear twice or disappear between stages.

The federal guidance notes recordkeeping responsibilities and expects employers to know which selection procedures are being used on their behalf. That matters when sourcing, screening, or assessments involve a service provider. Delegating work does not eliminate the employer's need to understand the procedure or its results.

Use stage analysis as a fault-location method

When the complete selection process produces a disparity signal, stage comparisons can narrow the investigation. A team might see similar rates through minimum-qualification review, followed by a sharp difference at an assessment or manager screen. That pattern tells the employer where to examine job relatedness, administration, missing data, and alternatives.

The inverse can also occur. A large stage difference may be offset later, leaving no overall difference in the period. The EEOC guidance explains when component analysis is expected under the Uniform Guidelines, but employers should get qualified advice before deciding what a particular result requires. Operational analysts should present the numbers and definitions without attaching a legal verdict.

The EEOC's overview of employment tests and selection procedures states that a neutral procedure can be unlawful if it disproportionately excludes people in a protected group and is not job related and consistent with business necessity. Its focus on tests is broad enough to include many screens used in hiring. OPM's assessment resources reinforce the importance of matching an assessment to job requirements.

Small cohorts need context

Ratios can move sharply when there are few decisions. If one group has five candidates, a single outcome changes its selection rate by twenty percentage points. A four-fifths calculation still produces a number, but that number may be unstable.

The federal questions and answers discuss circumstances in which small numbers make the four-fifths rule inappropriate, and circumstances involving larger numbers where smaller differences may matter. This is why a dashboard should display counts beside rates. It should also preserve time windows rather than silently pooling periods until a preferred answer appears.

Useful review packets show the numerator, denominator, impact ratio, missing demographic response, job, location, procedure version, period, and any operational interruption. They do not rank managers by a raw ratio or expose applicant demographic data to everyone who can view a recruiting dashboard. Access and purpose limitations belong in the design.

Data Sources and Methodology

We compared the calculation and process language in the eCFR text with explanations from the EEOC and U.S. Department of Justice. We used OPM material only to frame the relationship between job requirements and assessment choice. The analysis then identified the minimum recruiting events needed to reproduce the calculations. Additional checks used OPM job analysis, OPM structured interview guidance, OPM work-sample guidance, EEOC systemic enforcement material, and the EEOC Compliance Manual for scope.

No employer data were analyzed. The numerical example is illustrative and does not describe StopHighTurnover, a client, or a market benchmark. We did not choose a statistical significance test, establish a legal review protocol, or address every federal, state, and local requirement.

Limits on interpretation

The four-fifths rule can draw attention to a difference, but a ratio cannot show intent, validity, business necessity, applicant availability, or the reason for an outcome. Missing self-identification data and inconsistent disposition rules can distort the calculation. Combining jobs or periods can hide meaningful variation.

These materials are general research, not legal advice. Employers should determine applicable obligations with qualified counsel and appropriate specialists. Recruiting operations can protect event definitions and assemble reviewable records, but it should not make legal findings or alter selection criteria without authorization.

Evidence-led conclusion

Hiring teams need both overall and stage-level views, used for different purposes. The complete selection process is the proper starting point in the federal guidance. Stage analysis helps locate where a disparity may have arisen. The analysis is only as sound as its cohort, event definitions, and records. A four-fifths result should open a careful review, not close one with an automatic verdict.

Sources

1. Uniform Guidelines on Employee Selection Procedures, Electronic Code of Federal Regulations.
2. Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines, EEOC.
3. Employment Tests and Selection Procedures, EEOC.
4. Uniform Guidelines Questions and Answers, U.S. Department of Justice.
5. Assessment and Selection, U.S. Office of Personnel Management.