Hiring Discrimination and the Task Content of Jobs: Evidence from a Large-Scale Résumé Audit

Abstract: We conducted a large-scale resume audit of 36,880 applications to 9,220 job advertisements for new college graduates across the United States. Firms express task preferences through job-advertisement text, which we link to occupation-level task measures from O*NET and the American Community Survey. We develop a model in which discrimination increases with evaluative discretion, defined as the share of hiring decisions driven by subjective rather than verifiable assessment. Callback gaps vary systematically with the task content of jobs. In management occupations, callbacks are 28 to 43 percent lower for Black men, Black women, White women, and Hispanic men than for otherwise identical White men. Broad occupation categories conceal important variation in task demands. When jobs are grouped by task intensity, discrimination concentrates in positions combining high analytical and interpersonal demands with low routine content. Decomposing task content into subjective-evaluation and objective-precision components, we find that subjective evaluation widens callback gaps while objective precision compresses them. Customer contact amplifies this divergence, widening gaps in non-routine jobs but not in routine jobs. Randomly assigned resume credentials that increase callbacks on average reduce gaps in low-discretion jobs but not in high-discretion jobs. Early-career exclusion from high-return task bundles may entrench long-run demographic gaps in employment outcomes.

• The paper finds that discrimination is most pronounced in roles with high analytical and interpersonal demands, with callback rates up to 43% lower for minority applicants.
• The paper employs a randomized résumé audit linking detailed task indices from O*NET and ACS to isolate the role of subjective evaluation discretion.
• The paper reveals that verifiable credentials reduce discrimination in routine jobs but fail in high-discretion roles, suggesting targeted policy interventions.

Task Content and Hiring Discrimination: An Audit-Based Evaluation

Experimental Design and Context

This paper presents a large-scale résumé audit of 36,880 applications submitted to 9,220 job advertisements for new college graduates across the U.S., focusing on occupational and task content as key moderators of hiring discrimination. The audit randomized the race and gender signaled by applicant names across racial groups (White, Black, Hispanic) and genders, while ensuring random assignment of résumé credentials such as internships and programming experience. Firm preferences for tasks, as expressed in job-advertisement text, were linked to detailed occupation-level task measures from O*NET and ACS.

Crucially, the experiment isolates demand-side discrimination at the callback stage—free from confounding supply-side applicant self-selection—by examining detailed heterogeneity across occupation and task contexts. By focusing on new college graduates, the design allows precise identification of discrimination at labor market entry, avoiding the complications of previous work experience or complex work histories.

Theoretical Framework: Evaluative Discretion

The paper develops a model of discrimination that centers on evaluative discretion, defined as the proportion of hiring decisions reliant on subjective rather than verifiable assessment (i.e., the share of signal variance contributed by unstructured subjective evaluation in the hiring process). Two constructs represent task effects on signal variance: subjective evaluation noise (rising with analytical and interpersonal intensity), and objective evaluation precision (enhanced by routine task intensity).

The sufficient statistic for predicted discrimination is the discretion index: $E_j^* = \frac{B_j}{B_j + P_j},$ where $B_j$ encodes subjective noise (analytical/interpersonal intensity) and $P_j$ objective precision (routine cognitive intensity). The callback rate gap between majority and minority applicants scales positively with $E_j^*$. The model predicts that discrimination is largest for job types with high analytical and interpersonal demands and low routine content, and that customer contact will amplify discrimination only in jobs with high evaluative discretion.

Results: Heterogeneity across Occupation and Task Profiles

1. Occupational Group Differences.

Discrimination varies sharply by occupation. In management roles, callbacks for Black men, Black women, White women, and Hispanic men are 28–43% lower than for identically credentialed White men. Black men face a 5.1 percentage point callback gap in management but only 1.7 points in office/admin support. Gaps are negligible for Hispanic women across all categories.

2. Task Bundle Stratification.

K-means clustering of jobs by task content confirms that discrimination concentrates in occupations with high analytical and interpersonal intensity and low routine content, regardless of formal occupation group. There is minimal discrimination in jobs with high routine (and thus objective) evaluative components. Task profile–driven differences persist after controlling for job-ad fixed effects and are robust to alternative clustering, direct task-interaction specifications, and text-derived ad-level measures.

3. Evaluative Discretion and Decomposition.

Regression decompositions using proxies for subjective noise ($\hat{B}$) and objective precision ($\hat{P}$) demonstrate that the subjective component widens callback gaps, whereas the objective/routine component compresses them—the subjective–objective gap is statistically significant and negative across minority groups, especially in high-customer-contact work ($-4.3$ percentage points, $p<0.01$). This pattern is inconsistent with a generic account in which job complexity per se increases discrimination; only evaluative discretion (subjectivity) does.

4. Credential Signal Attenuation.

Randomly assigned résumé credentials (e.g., social internships, programming skills) that improve callback rates overall only reduce minority callback gaps in low-discretion (objective, structured) jobs. In high-discretion jobs, high-return credentials provide no differential benefit. This effect is credential-specific: placebo credentials with no average callback return show no moderation by discretion level.

5. Customer Contact Effects.

Customer contact amplifies discrimination—but only in jobs with high analytical or interpersonal demands. Contact does not affect callback gaps in routine jobs. This result contravenes standard customer-prejudice models, which would predict discrimination to increase with contact intensity independent of task complexity.

Implications and Future Directions

Economic and Labor Market Impact

The results indicate that hiring discrimination is not uniform but is sharply stratified by job task content. Exclusion at the screening stage from occupations rich in analytical and interpersonal content—which offer higher returns and are gateways to management—likely entrenches long-run demographic gaps. Because early-career experience in such roles compounds into persistent wage differentials, discrimination in task allocation (mediated by subjective evaluation) likely acts as a structural mechanism propagating downstream inequality.

Theoretical Consequences

The empirical findings directly support the evaluative discretion model. Screening procedures relying on highly subjective evaluation create more room for differential treatment of minorities, despite identical credentials. The demonstrated attenuation of gaps by objective/structured evaluation points to the salience of algorithmic and rule-based filtering as a potential lever to mitigate discrimination.

The distinction between subjective noise and objective precision in evaluative protocols offers guidance for the design of more equitable hiring systems. It also challenges standard screening-difficulty or "customer-prejudice-only" frameworks, necessitating richer models that take into account the multidimensional structure of task demands and screening noisiness.

Prospects for Automation and Artificial Intelligence

Given the findings, one significant implication for AI-based selection systems is that increasing the objective and transparent component of evaluation (for example, by leveraging structured assessment tools, automated skill tests, or task-based signals) might narrow demographic callback gaps—provided these systems are engineered to be free of embedded biases. However, to the extent that organizations migrate subjective evaluation into less-structured interfaces (e.g., algorithmically filtered but manager-judgment–dominated selection), task-conditional patterns of discrimination may persist.

Potential future research could empirically measure the extent to which automation of high-discretion screening processes reduces or perhaps inadvertently shifts discrimination across dimensions not captured by the current discretion index. The theoretical framework and audit protocols in this paper establish a modular foundation for such investigations.

Conclusion

This study provides definitive evidence—using experimental variation and detailed task profiling—that hiring discrimination for entry-level college graduates is highly concentrated in jobs with high analytical and interpersonal demands and low routine structure. Discrimination increases with the subjective component of hiring evaluation (evaluative discretion) and is mitigated by structured, objective evaluative content. Credentials and signals that increase callback rates do not reduce gaps in high-discretion roles. These patterns have substantial implications for the perpetuation of racial and gender inequality in the labor market and for the design of interventions—algorithmic or otherwise—aimed at reducing occupational access gaps and improving equity in labor market outcomes.

Reference:

"Hiring Discrimination and the Task Content of Jobs: Evidence from a Large-Scale Résumé Audit" (2604.01933).