--- title: "The Weight a Neutral Criterion Carries: A Company Reading Its Own Filters" description: "Disparate impact occurs when a facially neutral criterion, applied uniformly, screens out particular groups at disproportionate rates. The effect arises not in the wording of the rule but where the rule meets the real distribution of the applicant pool, which means it becomes visible only when pass rates are measured criterion by criterion, and defensible only when the criterion's link to job performance can be demonstrated." url: https://www.beirek.com/en/blog/disparate-impact-neutral-criteria canonical: https://www.beirek.com/en/blog/disparate-impact-neutral-criteria published: 2025-04-26 modified: 2025-04-26 category: "Organisational Psychology" category_url: https://www.beirek.com/en/blog/category/organisational-psychology language: en-US reading_time_minutes: 8 publisher: BEIREK LLC publisher_url: https://www.beirek.com license: "© BEIREK LLC — citation with attribution and link permitted" keywords: ["disparate impact","hiring criteria design","supplier prequalification thresholds","job-relatedness justification","employment practices due diligence","candidate pool narrowing"] topics: ["Organisational decision architecture","Screening criteria and proxy variables","Recruitment economics and time-to-fill","Employment practices in transaction diligence","Contractor and supplier prequalification"] alternate_language_url: https://www.beirek.com/tr/blog/disparate-impact-neutral-criteria --- # The Weight a Neutral Criterion Carries: A Company Reading Its Own Filters > **In short:** Disparate impact occurs when a facially neutral criterion, applied uniformly, screens out particular groups at disproportionate rates. The effect arises not in the wording of the rule but where the rule meets the real distribution of the applicant pool, which means it becomes visible only when pass rates are measured criterion by criterion, and defensible only when the criterion's link to job performance can be demonstrated. *A hiring threshold, a promotion requirement or a supplier prequalification rule may carry no discriminatory intent whatsoever and still remove particular groups from consideration at systematically higher rates, because the effect is produced not in the drafting of the criterion but at the point where the criterion intersects the actual distribution of the population it is applied to — and the institutional cost that follows is far more often operational than legal.* --- In a hiring panel convened to settle the specification for technical roles, the discussion around a five-year requirement for uninterrupted sector experience turns almost entirely on the number of years, and almost never on who the requirement removes. The same pattern governs the setting of a minimum turnover threshold in a supplier prequalification list, where the debate concerns the level of the threshold rather than the composition of the supplier set that falls below it. In both rooms the criterion has been written in purely functional language, referring to no group whatsoever — no mention of gender, age, geography or institutional origin appears anywhere in the text. Once applied, however, the criterion produces pass rates that are not evenly distributed across the population: a requirement for uninterrupted experience removes those who have stepped away from a career at a markedly higher rate, while a turnover threshold does the same to suppliers who are newly established or who work against a single concentrated customer. This asymmetry is generated not in the drafting of the rule but at the point where the rule meets a real-world distribution. Whoever designs a criterion sees the internal logic of that criterion rather than the way the attribute it measures is spread across the population, and so long as the logic remains internally coherent it goes unexamined. Most screening thresholds, moreover, are derived not from a stated need but from the threshold of the preceding cycle; five years is required this year because five years was required last year, and the specific operational justification that produced the number in the first instance has long since disappeared from institutional memory. The name for this pattern is disparate impact — the disproportionate disadvantaging of particular groups by a rule that is neutral on its face and neutral in its application. What distinguishes the mechanism is that it removes intent from the analysis altogether: the state of mind of the person who set the threshold does not alter the outcome the threshold produces, and good faith therefore constitutes no defence. The effect emerges from the gap between what a criterion is meant to measure and what it in fact measures; a requirement for uninterrupted experience, intended as a proxy for competence, actually measures career continuity, while a turnover threshold, intended as a proxy for execution capacity, actually measures a company's age and the breadth of its customer base. This drift between proxy variable and target variable persists regardless of how reasonable the criterion appears on the page. Reading the tendency as an error would be misleading. A screening criterion is a shortcut that lowers the cost of deciding, and under certain conditions it is entirely rational: so long as the cost of assessing several hundred applications individually exceeds the cost of the errors a roughly calibrated filter produces, the filter earns its place. The difficulty lies not in the shortcut itself but in its persistence after the conditions under which it was calibrated have changed. When the composition of the applicant pool shifts, when the content of the role shifts, or when the correspondence between the measured attribute and actual performance weakens, the filter stops reducing cost and merely narrows the pool — and the section of the pool being narrowed remains invisible for as long as it goes unmeasured. The first layer of institutional cost is operational rather than legal, and it accumulates directly in the economics of recruitment. A criterion that structurally narrows the candidate pool reduces the number of qualified applications per open role, which lengthens time-to-fill, and each additional week enters the accounts either as the direct cost of a vacant position or as overtime absorbed by the existing team. The second effect of a narrowed pool sits on the compensation side: as the set of candidates clearing the filter contracts, the premium paid to that set rises, and the company reads a scarcity of its own making as a scarcity in the market. The same mechanism operates on the procurement side, where a prequalification threshold that thins the supplier pool weakens tender competition and leaves the resulting price above what a broader field would have produced. The second layer surfaces when parties concerned with the ownership of the company take their seats at the table. Where a diligence process requests pass rates for hiring and promotion criteria broken down by group, most companies discover that the data has never been collected — and the absence of data is priced on the buy side as uncertainty. That uncertainty is typically answered not through a headline discount but through the architecture of the transaction itself: representations and warranties extended to cover employment practices, an increased escrow proportion, or an independent practices review imposed as a condition precedent to closing. For companies selling into corporate customers the same gap appears at the supplier audit stage, where the consequence takes the more immediate form of a delayed contract or removal from an approved vendor list. The third layer accumulates most slowly and is the hardest to reverse. As a criterion removes the same section of the population year after year, the internal population of the company converges on an increasingly homogeneous profile, and that population then reproduces the following cycle's criteria in a form that rewards candidates resembling itself. The loop is self-sustaining, because the evidence offered for the validity of the criterion is the performance of current employees — who are, by construction, precisely those who passed through it, while no observation whatsoever has been generated about how those who did not pass would have performed. For as long as a company measures the success of its own filter by looking only at those the filter admitted, the cost of the filter remains unobservable. What neutralises this tendency is not individual awareness; greater care on the part of whoever drafts a criterion does not change where that criterion intersects the distribution. Neutralisation requires a design discipline composed of three separate elements. The first is a written justification of job-relatedness for every screening criterion, recording the link between the attribute the criterion measures and the competence the role actually demands at the moment the criterion is set; where no justification can be written, the criterion was already unnecessary. The second is the measurement of pass rates at the level of each individual criterion — how many candidates each filter removes, and the composition of those removed, captured within the decision flow rather than reconstructed afterwards. The third is a systematic search for an alternative that measures the same competence in a less restrictive way: demonstrable output in place of uninterrupted tenure, evidence of execution on a reference project in place of a turnover threshold, structured technical assessment in place of a credential requirement. In the capital-intensive projects it manages, BEIREK places this discipline at the prequalification and team formation stage. When contractor and supplier selection criteria are established, each threshold — turnover, number of reference projects, certification, insurance limits — is entered into a criterion justification record against the specific execution risk it addresses, and any threshold whose justification cannot be traced back to the project risk matrix is removed from the list. Once the prequalification round closes, the suppliers screened out are reported line by line against the threshold that removed them, and where a single threshold is found to have eliminated a distinct section of the pool on its own, that threshold is recalibrated before the tender opens or replaced with a compensating security mechanism. The same recording logic is applied to the project organisation's own staffing. Competence definitions for key positions are derived from the work breakdown structure of the scope rather than from the staffing profile of previous projects, and a one-to-one correspondence is established between the position definition and the candidate assessment form, so that no criterion lacking a counterpart on the form can be used to screen. In assessment sessions the counter-argument role is assigned explicitly to a named individual whose sole function is to interrogate which criterion removed each rejected candidate — with the result that the rationale for exclusion is recorded simultaneously with the decision rather than reconstructed later. What these mechanisms share is the placement of the intervention at the moment the criterion is designed rather than at the moment a decision is taken. Correcting the effect of a criterion after it has entered force requires retrospective data and makes the correction itself look arbitrary; where the justification was written and a less restrictive alternative considered at the point of design, the decision rests on defensible ground whatever the resulting distribution turns out to be. Documentation does not guarantee an outcome; it shows how the decision was constructed, and that is precisely what a review table asks for. A company's filters ultimately describe what it is rather than what it intends to be. The question nobody raises while criteria are being drafted — who does this threshold remove, and is it true that those removed could not do this work — tends to be asked years later, either when no candidate can be found or when someone else is sitting across the table. The difference between a company that puts that question to itself at the moment the criterion is set and a company that searches for an answer once another party has asked it amounts, in the end, to nothing more than a discipline of record. ## Key Points - A criterion drafted in entirely neutral terms can still screen out particular groups systematically, because the effect originates in the distribution of the pool rather than in the intent of the drafter. - The legitimacy of a screening threshold rests not on how reasonable it appears but on a demonstrable link between what it measures and what the role actually requires. - The first institutional cost of disparate impact is rarely litigation; it is a narrowed candidate pool, a longer time-to-fill and a wage premium the company reads as market scarcity rather than as a product of its own filter. - Because the effect is created when the criterion is designed rather than when a decision is made, any neutralising mechanism has to be placed at the design stage. - Absent a measurement regime that records pass rates at the level of each individual threshold, a company never learns what its own filters are removing. ## Questions ### How does disparate impact differ from direct discrimination? Direct discrimination involves a practice that explicitly targets or differently treats a group, and it carries an element of intent. Disparate impact is independent of intent: the rule applies uniformly, refers to no group, and yet removes particular groups at disproportionate rates because of the structure of the population it meets. The distinguishing test is not the state of mind of the drafter but the distribution of pass rates the criterion produces. ### How can a company tell whether a hiring criterion produces disproportionate effect? Only through outcome data. For each screening criterion, the number of applicants, the number clearing that specific threshold and the composition of those removed must be recorded at criterion level. Looking at aggregate hiring numbers is insufficient, because the effect may concentrate in a single threshold and dilute in the total. Where measurement is not embedded within the decision flow, what the filter removes is never observable. ### Should thresholds such as years of experience or minimum turnover be abandoned? They can be retained; the difficulty lies not in the existence of a threshold but in the demonstrability of its rationale. A threshold remains defensible so long as the link between what it measures and what the work actually requires has been written down, and so long as an alternative measuring the same competence in a less restrictive manner has been considered. Thresholds justified only by the practice of the preceding cycle merely narrow the pool. ### How does this risk surface during a sale or investment process? Diligence typically requests pass rates for hiring, promotion and supplier selection criteria. Where the data has never been collected, the absence is priced as uncertainty, and the answer usually appears in the transaction structure rather than the price: representations and warranties extended to cover employment practices, a higher escrow proportion, or an independent practices review imposed as a condition precedent to closing. --- Source: https://www.beirek.com/en/blog/disparate-impact-neutral-criteria Publisher: BEIREK LLC — https://www.beirek.com