# Getting It Right: DEI and qualifications

Research date: 6 October 2026. Scope: public, general-audience factual education, primarily U.S. hiring, with explicitly identified international comparisons. This is a targeted evidence review, not a preregistered systematic review or an individual employment assessment.

## Editorial finding

**Recommended headline: “A DEI label does not show that someone is unqualified.”**

Alternative: **“Fair hiring and high standards can go together.”**

The defensible correction is that a program label, or someone's demographic identity, cannot establish that they are unqualified. Evidence strongly supports a second point: equally credentialed applicants can receive different treatment. Evidence for remedies is intervention-specific and mixed. Do not replace “DEI always lowers standards” with “DEI always selects the best person.”

The proposed claim that DEI prevents less-qualified advantaged people from taking jobs from more-qualified disadvantaged people is **a possible mechanism, not an established general result**. To substantiate it for a particular policy requires a credible comparison of selection with and without that policy, actual job-relevant qualifications or performance, and evidence about the affected candidates. Most studies below measure callbacks, applications, or representation rather than comparative job performance.

## Short explanatory page draft

### A DEI label does not show that someone is unqualified

“DEI” stands for diversity, equity and inclusion. It covers different practices, not one hiring rule. Expanding recruitment, using consistent interview questions, improving accessibility, providing training, and giving demographic preferences are different actions. They need different evaluations. Even the U.S. Equal Employment Opportunity Commission describes DEI as a broad term, not a category defined in Title VII. [EEOC](https://www.eeoc.gov/wysk/what-you-should-know-about-dei-related-discrimination-work)

There is good evidence that hiring is not always based only on credentials. A U.S. meta-analysis found that white applicants received 36% more callbacks than equally qualified Black applicants across studies conducted since 1989. That is a relative difference in callbacks, not 36 percentage points, and it describes the historical studies, not every employer today. [Quillian and colleagues, 2017](https://www.russellsage.org/sites/default/files/10870.full_.pdf)

That matters because a qualified person can be overlooked before anyone assesses how well they could do the job. Clear, job-related standards and consistent scoring can make assessment more reliable. Structured interviews ask comparable questions and apply common scoring criteria; research supports their value for predicting performance. [OPM](https://www.opm.gov/policy-data-oversight/assessment-and-selection/other-assessment-methods/structured-interviews/), [Sackett and colleagues, 2022](https://pubmed.ncbi.nlm.nih.gov/34968080/)

But a good intention is not proof of a good result. A large randomized study of a brief diversity training found mostly limited behavioral effects. In a French experiment, anonymizing résumés actually reduced minority applicants’ opportunities at participating firms. Programs should be evaluated by what they do and what happens afterward. [Chang and colleagues, 2019](https://pmc.ncbi.nlm.nih.gov/articles/PMC6475398/), [Behaghel and colleagues, 2015](https://doi.org/10.1257/app.20140185)

The useful questions are: What does this job require? How were candidates assessed? Were those standards applied consistently? Did the process identify people who perform well?

Someone’s race, sex, or association with a DEI program cannot answer those questions. Claims about their qualifications need evidence about their qualifications.

## Claim-level confidence and limits

| Claim | Judgment | Reason |
|---|---|---|
| DEI means unqualified | Unsupported generalization | Neither a broad program label nor identity establishes job competence. |
| Matched credentials always receive equal consideration | Contradicted in studied settings | Randomized correspondence tests isolate effects of identity signals on employer responses. |
| Bias removal can improve access for qualified applicants without lowering a minimum standard | Defensible mechanism | Access and assessment procedures can change while job requirements stay fixed; effects still require evaluation. |
| Every DEI policy selects higher-performing employees | Not established | The interventions, comparison conditions, outcomes and evidence quality vary. |
| Some affirmative-action hires have lower measured credentials | Supported in some older observational evidence | Does not equate to failing minimum qualifications, nor necessarily worse performance. |
| Representation gains prove improved merit or prove lower standards | Neither inference is justified | Composition is a different outcome from qualifications and performance. |
| White men are generally less qualified or undeserving | Unsupported and inappropriate | Population evidence cannot establish an individual's merit; gender effects also vary by job and employer. |

## Evidence matrix

Full-text access means the complete article or manuscript was accessible through the research tool; relevant methods/results/limitations were inspected. It does not mean data or code were independently reproduced. “Abstract only” is explicitly weaker review depth.

| Study/source | Design, scope and result | Quality and inference limits | Access |
|---|---|---|---|
| Quillian, Pager, Hexel & Midtbøen (2017), PNAS, DOI 10.1073/pnas.1706255114. [Paper](https://www.russellsage.org/sites/default/files/10870.full_.pdf) | Meta-analysis: 28 U.S. field experiments, 55,842 applications, 26,326 positions. Main post-1989 analysis: 24 studies. White applicants received 36% more callbacks than Black applicants (95% CI 25–47%) and 24% more than Latino applicants (15–33%). | Strong evidence of unequal treatment among experimentally comparable applications. Included unpublished studies. No direct estimate of modern DEI effectiveness, actual worker performance, or universal employer behavior. Name signals and sampled occupations constrain generalization. | Full text; PMC initially returned a CAPTCHA page, publisher-linked funder PDF accessible. |
| Quillian & Lee (2023), PNAS 120(6), e2212875120. [Paper](https://pmc.ncbi.nlm.nih.gov/articles/PMC9963383/) | Meta-analysis of 90 field studies, over 170,000 applications, six Western countries. Generally stable discrimination over studied decades, with country/group variation including a decline in France. | Country-specific composition and study controls matter. Different time windows do not constitute a 2026 labor-market census. Trend stability is distinct from proof that interventions are ineffective: counterfactual trends are not identified. | Indexed full article text was returned by search; direct open hit CAPTCHA. Treat as indexed-text inspection rather than successful direct full-text open. |
| Kline, Rose & Walters (2022), QJE 137(4), 1963–2036. [Published abstract](https://academic.oup.com/qje/article-abstract/137/4/1963/6605934) | Randomized correspondence experiment: over 83,000 applications to 108 large U.S. employers. Distinctively Black names lowered employer contact by 2.1 percentage points. Average gender gap insignificant; substantial employer variation in both directions. | Strong internal validity for contact-stage identity signals. Not a random sample of all employers; no job-performance outcome. Correlations with centralized recruiting are not randomized tests of centralization. | Published abstract inspected; NBER and author PDF retrieval failed. |
| Lippens, Vermeiren & Baert (2023), European Economic Review 151, 104315. [Full paper](https://backoffice.biblio.ugent.be/download/8771122/8773148) | Broad meta-analysis of correspondence experiments published 2005–2020, covering nearly 965,000 applications and multiple discrimination grounds. Gender findings highly heterogeneous; pooled female callback advantage about 4%, with discrimination against both men and women in individual settings. | Important check against treating every minority or every woman as uniformly disadvantaged in all hiring contexts. Pooled averages depend on occupation/sample mix; callbacks exclude pay, promotion and retention. Not independent of every other meta-analysis here because underlying studies overlap. | Full text. |
| Sackett, Zhang, Berry & Lievens (2022), Journal of Applied Psychology 107, 2040–2068. [Abstract](https://pubmed.ncbi.nlm.nih.gov/34968080/); [2023 author follow-up](https://www.cambridge.org/core/journals/industrial-and-organizational-psychology/article/revisiting-the-design-of-selection-systems-in-light-of-new-findings-regarding-the-validity-of-widely-used-predictors/A20984B138319E3D432E643978BF026D) | Re-examined selection-validity meta-analyses, correcting prior overcorrection for range restriction. Structured interviews ranked highest by mean validity; many previous estimates reduced. Follow-up reports substantial between-setting variation. | Predictive validity is not the same as a randomized causal effect on discrimination or an endorsement of every interview. Job analysis, criterion quality and implementation matter. Do not recycle old “85 years of research” coefficients without checking revisions. | 2022 abstract; 2023 full text. |
| OPM, Structured Interviews. [Official guidance](https://www.opm.gov/policy-data-oversight/assessment-and-selection/other-assessment-methods/structured-interviews/) | Describes predefined job-related questions, proficiency benchmarks, constrained discretion and increased rating agreement. | Operational description and synthesis, not a new trial. Supports an example of maintaining standards while changing assessment. Does not show all structured processes eliminate bias. | Full official page. |
| Flory, Leibbrandt, Rott & Stoddard (2021), Journal of Human Resources 56(1), 73–92. [Paper](https://research.vu.nl/ws/files/127554573/project_muse_780196_Increasing_Workplace_Diversity.pdf) | Natural field experiment varying recruitment messages at a major financial-services employer. Explicit diversity interest increased racial-minority interest and applications; selection estimates also increased but were imprecise. Gender effects were generally not significant. | Randomized message exposure supports effects of these messages in this setting. Undergraduate recruitment and limited minority sample constrain generalization. The raw selection change (4% to 12%) had p = 0.25; selection was for a recruiting program, not employment or productivity. Recruitment is not interchangeable with a hiring preference. | Full text. |
| Chang et al. (2019), PNAS 116(16), 7778–7783. [Paper](https://pmc.ncbi.nlm.nih.gov/articles/PMC6475398/) | Preregistered randomized experiment, 3,016 employees in one global organization, active placebo, behavior followed up to 20 weeks. Attitude changes and some subgroup behavioral changes, mostly null behavioral results overall. | Stronger causal design than before/after surveys. Voluntary participation and one organization limit external validity. Immediate attitude responses may reflect demand effects. No evidence of backlash in this trial; do not describe it as proving training is harmful. | Full text. |
| Behaghel, Crépon & Le Barbanchon (2015), AEJ: Applied Economics 7(3), 1–27. [Published abstract](https://doi.org/10.1257/app.20140185); [author manuscript](https://www.parisschoolofeconomics.com/behaghel-luc/cva.pdf) | French public employment-service experiment randomized anonymous versus identified résumés within volunteering firms. Minority applicants were less likely to be interviewed and hired under anonymity. Authors analyze firm self-selection and loss of contextual interpretation of résumé signals. | Credible adverse effect in the participating-firm population. Firms self-selected into the pilot; cannot infer all anonymous hiring is harmful. Masking can remove helpful as well as prejudicial information. | Published abstract and full 2014 author manuscript; final article methods not separately compared line by line. |
| Miller (2017), AEJ: Applied Economics 9(3), 152–190. [Paper](https://conrad-miller.github.io/app.20160121.pdf) | Event-study design using federal-contractor regulation/deregulation, 1978–2004 administrative data. Black employment share increased about 0.8 percentage points in five years after initial regulation and continued growing after deregulation. | Quasi-experimental, contingent on identifying assumptions; contracting changes are not randomized. Improved screening is a proposed mechanism with suggestive evidence, not direct proof. Does not measure the relative competence of displaced hires. Historical policy, not a description of current contractor duties. | Full text. |
| Holzer & Neumark (1999), Journal of Labor Economics 17(3), 534–569. [Published abstract](https://www.journals.uchicago.edu/doi/10.1086/209930) | Employer–employee data from four U.S. metro areas. Affirmative-action hires in studied groups sometimes had lower educational credentials, but most groups did not show weaker measured job performance. | Directly relevant distinction between credentials and performance, but observational, old and context-limited. Employer policies and applicants are not randomized; absence of detected differences is not equivalence proof. Avoid “all hires equally qualified.” | Published abstract; NBER manuscript retrieval failed. |
| Nichols, Axt, Gosnell & Ariely (2023), Nature Human Behaviour 7, 2212–2227. [Author institution abstract](https://scholars.duke.edu/publication/1605067); DOI 10.1038/s41562-023-01731-5 | Field experiment, 31,928 website visitors and 1,585 applicants, manipulating workforce images. Little evidence of effects on applicant demographics or quality; application volumes varied with combinations of displayed diversity. | Subtle imagery is a specific recruitment cue, not a test of the entire DEI umbrella. Application quality ratings do not establish eventual job performance. | Institutional abstract; direct PMC full-text access blocked by CAPTCHA. |
| Leibbrandt & List (2018), NBER Working Paper 25035. [Abstract](https://www.nber.org/papers/w25035) | Natural field experiment across 10 U.S. cities and nearly 2,500 job seekers. An equal-opportunity statement reduced minority willingness to apply in that experiment. | Working-paper evidence; message framing and recruitment context matter. Do not generalize to all EEO statements or assume a universal stigma mechanism. Included as a contrary result, not a principal page citation. | Search-indexed NBER abstract only; direct open failed. |
| Zhao, Hsieh, Faulkner & Smith (2025), A systematic meta-review of organizational diversity and inclusion interventions and their associated outcomes, Equality, Diversity and Inclusion 44(9), 53–71, DOI 10.1108/EDI-02-2024-0085. [Full paper](https://doi.org/10.1108/EDI-02-2024-0085) | Review of 37 reviews, 12 intervention categories and 22 outcomes. No intervention category uniformly superior; training evidence mainly concerns awareness/learning, with mixed recruitment-policy findings. | Published recently but database search updated only through November 2021. Mixed review designs, overlapping primary studies and uneven population coverage. Useful scope map, not a pooled causal effect of DEI. | Full text, methods and limitations inspected. |

## Definitions and causal distinctions

- **Qualified versus most qualified:** Meeting required competencies is not identical to being top-ranked by a particular test. A lower credential score need not mean below the job threshold; neither does it automatically mean the score is irrelevant.
- **Credentials versus performance:** Degrees, experience and tests are measures or predictors. Their predictive validity can be examined. Job performance itself may be measured noisily or through biased supervisor ratings.
- **Opportunity versus selection:** Reaching more candidates changes who knows about and applies for a job. Standardized scoring changes how applicants are assessed. Training changes knowledge or behavior. Preferences change selection rules. The same label can contain several mechanisms.
- **Networks:** Referrals can supply useful information, while a narrow recruiting pool can miss capable candidates. The specific claim that advantaged networks systematically select less-qualified hires is not established by callback experiments. Describe it as a hypothesis requiring direct research, not an inference from group composition.
- **Fairness and utility:** Equal treatment is a normative/legal principle; representation, applicant quality, productivity and belonging are distinct empirical outcomes. A policy can improve one and not another.
- **Evidence about groups versus individuals:** Experimental average treatment effects are not evidence that any named employee lacks ability, was favored, or took a position from someone more qualified.

## Contrary findings and editorial safeguards

1. Explicitly retain anonymity's adverse result and training's limited effects. They are central to avoiding advocacy masquerading as evidence review.
2. Do not imply that all women are disadvantaged at the callback stage. The large employer study and broad meta-analysis reveal variation and advantages in some settings.
3. Do not treat representation as a productivity measure. Miller's persistent changes cannot establish which individual was best for a job.
4. Do not claim “lower credentials never happen.” Holzer–Neumark found some credential differences alongside much less evidence of performance differences.
5. Do not promise structured interviews remove all bias. Their questions, rubrics and performance criteria still need validation.
6. Do not translate a non-significant effect into proof of no effect. Report uncertainty and study scope.
7. Do not count overlapping studies across meta-analyses as independent replications.

## Narrative prominence: documented examples, not assumed virality

The White House's 22 January 2025 fact sheet explicitly frames DEI as conflicting with competence-based employment. This is an identifiable, high-level institutional example of the narrative. It does not establish how many Americans endorse it or how viral it is. The page can say “DEI is sometimes presented as the opposite of merit,” linked to the example, but should avoid “everyone believes,” “the dominant myth,” or invented reach statistics. [Primary source](https://www.whitehouse.gov/fact-sheets/2025/01/fact-sheet-president-donald-j-trump-protects-civil-rights-and-merit-based-opportunity-by-ending-illegal-dei/)

Avoid repeating unsupported allegations about aviation safety or attributing failures to named employees. A political statement is evidence that a statement was made, not evidence that its causal allegations are true.

## Narrow legal context, checked 6 October 2026

The EEOC's current posted technical assistance says DEI is not defined in Title VII and that employment actions motivated by protected characteristics may violate that law. It applies the prohibition to workers of all groups. This supports the narrow point that a DEI label is no automatic legal shield. It also does not establish that everything called DEI is illegal. Agency guidance is not a comprehensive account of statutes, court precedent or jurisdiction-specific obligations. The educational page need not adjudicate contested affirmative-action law. Do not turn this dossier into advice for making protected-characteristic-based individual employment decisions. [EEOC source](https://www.eeoc.gov/wysk/what-you-should-know-about-dei-related-discrimination-work)

## Search and inclusion audit

### Search method

Targeted public-web searches on 6 October 2026, using primary studies, author/institutional repositories, scholarly publishers, NBER and official government pages. Followed backward/forward leads from meta-analyses and deliberately searched for mixed, null and adverse effects. Search engines were used for discovery, not as proof beyond the indexed source text actually available. No private data were accessed. No databases were exhaustively exported; no PRISMA counts or claims of exhaustive coverage are warranted.

### Inclusion rules

- Prioritize actual hiring correspondence/audit experiments, randomized workplace/recruitment interventions, transparent quasi-experiments and systematic syntheses.
- Include observational evidence when directly about qualifications/performance, but downgrade causal conclusions.
- Distinguish U.S. from other countries and historical from current evidence.
- Retain disconfirming and null findings. Prefer measured behavior over self-reported intentions, without treating attitudes as meaningless.
- Use official documents to establish definitions, posted legal guidance and public rhetoric, never as substitutes for independent causal evidence.
- Exclude social-media anecdotes, commercial HR claims, unverified employer stories, generic corporate profitability correlations and studies that infer individual ability from demographics.
- A retrieved institutional record for “Insights from an updated personnel selection meta-analytic matrix” was marked “withdrawn”; it was excluded rather than used to strengthen the argument. [Record](https://ink.library.smu.edu.sg/lkcsb_research/7534/)

### Search log (queries issued, grouped in execution order)

1. “Quillian Pager Hexel Midtboen 2017 meta analysis field experiments hiring discrimination PNAS”; “Chang Milkman diversity training randomized field experiment 2019 PNAS”; “Behaghel Crepon Le Barbanchon unintended effects anonymous resumes 2015”.
2. “Sackett Zhang Berry Lievens 2022 revisiting meta analytic estimates validity personnel selection structured interviews”; “Miller 2017 persistent effect temporary affirmative action American Economic Journal federal contractors”; “site.nber.org diversity recruiting field experiment application qualifications”.
3. “site.whitehouse.gov 2025 DEI merit competence hiring”; “site.eeoc.gov what you should know DEI related discrimination 2025”; “site.nber.org Bertrand Mullainathan Emily Greg Lakisha Jamal”; “site.nber.org affirmative action hiring performance qualifications Holzer Neumark”.
4. “Kline Rose Walters 2022 systemic discrimination large US employers 2.1 percentage points”; “site.pnas.org diversity recruitment field experiment Flory Leibbrandt Rott Stoddard”; “site.opm.gov structured interviews same questions rating scale”.
5. “\"A field study of the impacts\" \"quality\" diversity recruitment 2023”; “\"Trends in racial and ethnic discrimination\" \"2023\" 90”; “\"Increasing Workplace Diversity\" \"Flory\" \"2021\"”; “\"Unintended Effects of Anonymous\" pdf Behaghel 2015 2014”.
6. “\"Trends in racial and ethnic discrimination in hiring in six Western countries\" 2023 Quillian Lee”; “\"Systemic Discrimination Among Large U.S. Employers\" pdf berkeley”; “\"Are Affirmative Action Hires Less Qualified\" \"performance\" education 1999”.
7. “diversity equity inclusion interventions workplace systematic review randomized trials 2024 2025 2026 effectiveness”; “gender discrimination hiring meta analysis correspondence experiments 2023 Lippens Vermeiren Baert”.

### Access and verification limitations

Some PMC pages returned CAPTCHA interstitials and some NBER/author PDF opens returned tool errors. No CAPTCHA was solved. Accessible publisher, funder and author-repository routes were used where possible, and partial access is marked per study. No unpublished effect-size reanalysis, systematic correction/retraction database audit, or comprehensive 2026 literature search was performed. The withdrawn repository record noted above was explicitly excluded; no inference about unrelated papers by those authors follows from that record.

### Before publication

Confirm the final paper text for abstract-only studies if adding precise secondary outcomes; verify citations still resolve; date the evidence; retain caveats in the visible page rather than burying all of them; do not promote the older single-study results into universal claims. A useful reader-facing source drawer can show design, sample, main outcome and the question the study does **not** answer.