Difficulty: Hard
Why can a high-performing model still be unsuitable for HR decisions?
A Accuracy always establishes fairness
B It may cause unfair impacts
C Models eliminate all judgment
D Historical data is always neutral
Answer: B. It may cause unfair impacts. Predictive performance alone does not establish fairness or suitability.
When should HR avoid using a personal attribute in a model?
A Whenever a field is numeric
B Only when data is old
C When dashboards are colorful
D When purpose or fairness fails
Answer: D. When purpose or fairness fails. Use requires a justified purpose and assessment of fairness and applicable rules.
How should analysts handle a proxy for a sensitive trait?
A Assess whether it reproduces bias
B Assume it is harmless
C Hide it from governance
D Use it because it predicts
Answer: A. Assess whether it reproduces bias. A proxy can reproduce sensitive-group differences even if the direct field is absent.
Why predefine an HR analysis question?
A Guarantee a favorable result
B Avoid sharing assumptions
C Replace all data-quality checks
D Limit scope and reduce fishing
Answer: D. Limit scope and reduce fishing. A clear question helps keep analysis focused and reduces selective interpretation.
What is survivorship bias in workforce data?
A All employee records are duplicated
B Only remaining employees are observed
C A survey response is anonymous
D A forecast has many scenarios
Answer: B. Only remaining employees are observed. Studying only employees who remain can omit information from those who left.
How may selection bias affect engagement surveys?
A Survey ratings become causal
B All comments are verified
C Respondents may differ systematically
D Response rates always equal
Answer: C. Respondents may differ systematically. Respondents may not represent employees who chose not to participate.
Why document analytical assumptions?
A Make models less accurate
B Avoid explaining limitations
C Enable review and replication
D Replace data governance
Answer: C. Enable review and replication. Assumption records let others review and reproduce the analysis.
Which step supports reproducible HR analysis?
A Version data and definitions
B Change calculations without notes
C Use private undocumented filters
D Delete prior results
Answer: A. Version data and definitions. Versioning helps establish which data and definitions produced each result.
What is model drift?
A Performance changes as conditions shift
B A missing employee record
C A dashboard color mismatch
D A change in survey wording
Answer: A. Performance changes as conditions shift. Model drift occurs when input or outcome patterns change and performance shifts.
How should HR respond to model drift?
A Trust the original validation forever
B Monitor, validate, and update
C Increase employee penalties
D Remove all human review
Answer: B. Monitor, validate, and update. Monitoring and revalidation can identify when models need adjustment.
Why test for disparate impact?
A Guarantee identical outcomes
B Remove all workforce differences
C Check unequal effects across groups
D Avoid examining model accuracy
Answer: C. Check unequal effects across groups. Impact testing can reveal whether a process affects groups differently.
What should a model card or documentation record?
A Employee passwords and names
B Only the final score
C Purpose, data, limits, evaluation
D Marketing claims alone
Answer: C. Purpose, data, limits, evaluation. Model documentation should describe intended use, data, evaluation, and limitations.
Who should own an HR analytics metric?
A Every employee equally
B No one after launch
C The chart designer only
D A named business steward
Answer: D. A named business steward. A named steward can maintain the definition and quality of a measure.
How can HR governance clarify accountability?
A Share all data by default
B Use undocumented approvals
C Remove audit history
D Assign owners and access roles
Answer: D. Assign owners and access roles. Clear owners and access roles support accountable data use.
Why separate analytics access from personnel decisions?
A Make analysis impossible
B Limit unnecessary identifiable access
C Guarantee a legal outcome
D Avoid reporting aggregate trends
Answer: B. Limit unnecessary identifiable access. Different access levels can reduce exposure while enabling appropriate analysis.
Which practice supports purpose limitation?
A Repurpose every dataset freely
B Keep data forever
C Share records with all managers
D Use data for stated purposes
Answer: D. Use data for stated purposes. Purpose limitation constrains information use to defined and appropriate purposes.
What should happen when data is no longer needed?
A Follow retention and deletion rules
B Keep copies without limit
C Publish anonymized names
D Move it to private email
Answer: A. Follow retention and deletion rules. Retention and deletion should follow applicable policy and requirements.
Why involve employees in HR data governance?
A Let them approve every metric
B Avoid explaining data use
C Replace privacy safeguards
D Surface concerns and build trust
Answer: D. Surface concerns and build trust. Employee input can reveal concerns about collection, interpretation, and use.
How should analytics teams respond to a misleading metric?
A Correct it and explain impact
B Leave it until next year
C Quietly change historical values
D Blame dashboard users
Answer: A. Correct it and explain impact. Transparent correction helps users understand how previous conclusions may be affected.
What does an audit trail record?
A Employee lunch preferences
B A manager’s informal opinion
C Data access and analytic changes
D Future turnover certainty
Answer: C. Data access and analytic changes. Audit trails can record access, changes, and analytical decisions.
Which is a responsible way to share a forecast?
A State uncertainty and use limits
B Call it certain prediction
C Rank employees by probability
D Hide affected populations
Answer: A. State uncertainty and use limits. Users should see uncertainty and understand appropriate limits on forecast use.
Why is human review important for HR model outputs?
A Humans always eliminate bias
B Context can change interpretation
C Review guarantees correct results
D Models cannot process data
Answer: B. Context can change interpretation. Human review adds context, though it still needs to be structured and fair.
How can HR analytics improve a process without profiling people?
A Analyze aggregate process bottlenecks
B Predict named employees’ loyalty
C Publish individual risk lists
D Infer protected characteristics
Answer: A. Analyze aggregate process bottlenecks. Process-level analysis can improve workflow without making individual predictions.
What is an appropriate response to a weak evidence base?
A Make a stronger claim
B Gather better data before concluding
C Use the most persuasive chart
D Treat estimates as facts
Answer: B. Gather better data before concluding. Weak evidence should lead to more data or cautious conclusions.
What should leaders ask before adopting an HR analytics recommendation?
A How impressive is the chart?
B Can it remove all exceptions?
C Evidence and risks behind it
D Will it guarantee retention?
Answer: C. Evidence and risks behind it. Decision-makers should examine evidence, limitations, and potential impacts.