| Course | MGT 475 Technology in Human Resource Management |
|---|---|
| Topic | Topic 4 |
| Paper type | Algorithmic screening audit |
| Length | About 1,000 words, 4 pages |
| Format | APA 7 student paper |
| School | Grand Canyon University (GCU) |
| Program | BS in Business Management |
| Updated | October 2026 |
Free sample paper for MGT 475 Topic 4
Trained on Ten Years of Mostly Male Managers: An Audit of an Algorithmic Screening Tool for Department Manager Hiring, From Training Data to an Impact Ratio of 0.64
[Author Name]
Colangelo College of Business, Grand Canyon University
MGT-475: Technology in Human Resource Management
Topic 4 Assignment
[Instructor Name]
[Date]
The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.
Trained on Ten Years of Mostly Male Managers: An Audit of an Algorithmic Screening Tool for Department Manager Hiring, From Training Data to an Impact Ratio of 0.64
What Is Being Audited
Prairie Market is piloting a vendor's screening tool for department manager openings in its stores. Applicants complete an online application, and the tool assigns each a score from 1 to 100 predicting success as a department manager. Applicants scoring above 60 advance to an interview with the store manager; the rest receive a rejection email. The tool has been used for six months in twelve stores. This audit asks where the tool's preferences come from, what inputs it uses, what outcomes it produces for different groups and whether anyone at the chain can explain an individual score. All figures are illustrative.
The Training Data
The audit starts with the training data, because a model of this kind learns what success looks like from past examples. The vendor built the model on ten years of Prairie Market's own department manager hires, using their performance ratings and tenure as the definition of success. Of those managers, 81 percent were men.
A model trained on that history will tend to favor applicants who resemble the people the chain hired before, whatever the reasons those people were hired. If the chain's past hiring favored men, for any reason, the model may reproduce that preference without anyone designing it to. Vendors of algorithmic hiring tools vary widely in what they disclose about how they build and validate their models and test them for bias (Raghavan et al., 2020).
The definition of success matters too. Performance ratings and tenure reflect the judgments of the managers who gave the ratings and the conditions that kept people in their jobs. If ratings were affected by bias, or if women managers left more often because of conditions at the chain, the model would learn those patterns as if they described ability.
The Inputs
The vendor disclosed most of the inputs the model uses. Two raise concerns.
Continuous employment history. The model rewards applicants without gaps in their work history. Gaps are more common among people who have taken time out for caregiving, a group that includes more women than men. A preference for continuous history may therefore disadvantage women for reasons unrelated to their ability to manage a department.
Distance from home to store. The model gives higher scores to applicants who live closer to the store, apparently because nearby managers stayed longer in the training data. Distance can stand in for neighborhood, and neighborhoods in the chain's markets differ in racial composition, so this input could disadvantage applicants of some races even though race is not an input.
Other inputs, such as years of retail supervisory experience and completion of a short scenario assessment, are more plainly related to the job.
The vendor declined to disclose two inputs, describing them as proprietary. The audit records that refusal. A tool whose inputs cannot all be examined cannot be fully assessed for bias, and the chain should not accept that limitation for a tool that decides who gets an interview.
The Outcomes
Table 1
Selection Rates by Sex at the Screening Stage
| Group | Applicants | Advanced to interview | Selection rate |
|---|---|---|---|
| Women | 300 | 42 | 14% |
| Men | 500 | 110 | 22% |
| Impact ratio | 14% / 22% = about 0.64 |
Table 1 shows who advanced to interview during the six-month pilot.
Under federal guidelines, when one group advances at less than 80 percent of the rate enjoyed by the best-placed group, the gap is generally read as evidence of adverse impact (Uniform Guidelines on Employee Selection Procedures, 2024). An impact ratio of about 0.64 falls well below that threshold. The Equal Employment Opportunity Commission has stated that employers using algorithmic tools for selection remain responsible for adverse impact those tools produce, even when the tool is designed or run by a vendor (U.S. Equal Employment Opportunity Commission, 2023).
The audit treats this result as a signal to investigate, not as proof that discrimination occurred. Adverse impact can be lawful if the selection procedure is job-related and consistent with business necessity, and if no less discriminatory alternative would serve the same purpose. But the chain cannot show that yet.
The audit also checked selection rates by race using applicants' voluntary self-identification. The numbers were too small in several groups to calculate reliable ratios, which is itself a reason to continue monitoring as more applicants pass through the tool.
Explaining Individual Scores
The audit asked three store managers and the talent acquisition director to explain why particular applicants scored below 60. None could. The vendor provides a score but no breakdown of which inputs drove it, and the contract does not require one. If a rejected applicant asked why, or if a regulator asked the chain to show that the tool measures job-related qualities, the chain would have no answer. A selection procedure that no one inside the organization can explain is difficult to defend and impossible to improve.
Explanation also matters for the chain's own learning. If a tool screens out applicants who later prove successful elsewhere in the chain, the only way to find out why is to see what drove their scores. Without explanations, the chain cannot correct the tool or decide when to stop using it.
Recommendations
The audit recommends four steps. First, suspend automatic rejection: every applicant who meets the posted minimum qualifications should be reviewed by a person until the issues below are resolved. Second, ask the vendor to remove the continuous employment and distance inputs, or to show evidence that they predict job performance and that no less discriminatory alternative would work as well. Third, require the vendor to provide, for each score, the main factors behind it, and to share its validation evidence. Fourth, rerun this audit after any change to the model, and quarterly thereafter, comparing selection rates by sex and by race. The tool may still prove useful, but only once the chain understands what it measures and can show that it measures the right things.
References
Raghavan, M., Barocas, S., Kleinberg, J., & Levy, K. (2020). Mitigating bias in algorithmic hiring: Evaluating claims and practices. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (pp. 469-481). Association for Computing Machinery. https://doi.org/10.1145/3351095.3372828
Uniform Guidelines on Employee Selection Procedures, 29 C.F.R. pt. 1607 (2024).
U.S. Equal Employment Opportunity Commission. (2023). Select issues: Assessing adverse impact in software, algorithms, and artificial intelligence used in employment selection procedures under Title VII of the Civil Rights Act of 1964. U.S. Equal Employment Opportunity Commission.
What the MGT 475 Topic 4 instructions ask for
The Topic 4 instructions for MGT-475 ask students to audit an automated tool used in employment decisions. Accepting a vendor's assurance, or concluding from one number that the tool discriminates, misses the analysis. The assignment rewards an audit that starts with the training data, examines inputs for proxies that may disadvantage protected groups, calculates selection rates and impact ratios correctly and interprets them carefully. Students should explain the employer's continuing responsibility under federal guidance, test whether anyone can explain individual results and recommend specific steps, such as suspending automatic rejection and requiring vendor disclosures, along with a schedule for repeating the audit.
How the MGT-475 Topic 4 example is put together
The audit opens with how the tool works and what it decides. The training data section explains how a mostly male history can shape a model's preferences and how the definition of success matters. The inputs section identifies employment continuity and distance as possible proxies and records the vendor's refusal to disclose two inputs. Table 1 shows selection rates and the impact ratio, followed by the four-fifths rule and federal guidance on vendor tools. A section reports that no one could explain individual scores. Four recommendations close the audit, including human review and quarterly reruns. The vendor's refusal to disclose two inputs is recorded as a finding in its own right.
Where the marks sit in the MGT 475 Topic 4 rubric
Instructors grading the screening audit reward an audit that looks at training data and inputs before outcomes. Strong papers identify inputs that may act as proxies for protected characteristics and explain why. Credit goes to calculating the impact ratio correctly and treating it as a signal to investigate rather than as proof. The rubric values recognizing the employer's responsibility for vendor tools and testing whether results can be explained. The highest band recommends concrete steps that protect applicants while the tool is examined, and a schedule for continued monitoring. Audits that rely on the vendor's claims or stop at a single number usually land in the lower bands. Recording what a vendor would not disclose is part of a complete audit.
MGT 475 Topic 4 help: mistakes that cost marks
Screening audits most often go wrong by accepting the vendor's framing that the tool is objective because it applies one rule to everyone. Consistency is what the tool delivers, and a paper that stops there has missed that a consistent rule learned from skewed history reproduces the skew at scale. Audits measuring outcomes at final hire dilute the tool's effect with every later human choice. Four-fifths results reported as proof of discrimination overstate the test, while results below the line dismissed as a statistical quirk understate it. Papers that never ask anyone at the employer to explain a score miss the accountability question the course is built around. Leaving the vendor responsible for the outcome misplaces it, since the employer that uses the tool generally answers for its effect on applicants.
Get an MGT-475 Topic 4 example written to your instructions
Send the MGT-475 Topic 4 instructions, the rubric posted in your classroom and any tool, vendor or applicant data the case describes. A custom example is written to those criteria, with training history examined, proxies flagged, selection rates computed where the tool acts, the four-fifths check applied as a signal and an owner named, in 24 to 48 hours. The first one is free.
More MGT 475 papers and BS in Business Management samples
- MGT 475 Topic 1: Hris Needs Analysis
- MGT 475 Topic 2: Hr Data Field Trace
- MGT 475 Topic 3: People Analytics Question Brief
- MGT 475 Topic 5: Recommendation Ownership Dq Post
- MGT 475 Topic 6: Employee Monitoring Assessment
- MGT 475 Topic 7: Hris Implementation Evaluation
- MGT 475 Topic 8: Hr Technology Oversight Plan
- BUS 317 Topic 4: Break-even Sensitivity Analysis
- MGT 470 Topic 4: Expatriate Package Analysis
- BUS 332 Topic 4: Service Failure Analysis
- MGT 465 Topic 4: Concerted Activity Analysis
MGT 475 Topic 4 questions, answered
Why does a tool trained on past hires reproduce past patterns?
Because it learns which applicant features went with success in the historical record, and that record reflects who managers chose. If most past department managers were men, features common among men become predictors whether or not they relate to the work. The tool then applies those associations to every applicant identically, which makes the pattern more consistent than any single manager ever was.
Is failing the four-fifths rule proof of discrimination?
No. Federal selection guidelines use the four-fifths comparison as a rule of thumb that flags a procedure for closer review. A lower ratio calls for investigation, including whether the procedure is job-related and whether a less discriminatory alternative exists. Small samples can also produce misleading ratios in either direction. The example is coursework analysis and draws no legal conclusion.
Are there laws aimed specifically at automated hiring tools?
A few, and more are proposed. New York City requires employers using automated employment decision tools to commission an independent bias audit and notify candidates, and Illinois regulates the use of artificial intelligence in analyzing video interviews. Title VII's disparate impact framework applies regardless of who built a tool. Rules change quickly here, so this is coursework context, not legal advice.
Where can I find a free MGT 475 Topic 4 sample paper?
The screening audit appears above in full, covering the training data, the inputs, the outcomes table, explanation of scores and the recommendations. Share the tool and the outcomes you are auditing, and a free audit can be written.
What is the impact ratio in MGT 475 Topic 4?
About 0.64. Women advanced at 14 percent and men at 22 percent, and 14 divided by 22 falls well below the four-fifths threshold used to flag adverse impact.