WORK6010 Chap.5 Staffing and Selection Analytics
Staffing and Selection Analytics
Selection analytics links applicant information, assessment results, hiring decisions and later job outcomes. A useful model must predict a job-relevant criterion and operate consistently across groups, stages and future cohorts. This chapter calculates within-group selection rates, explains criterion validity and distinguishes an observed adverse pattern from a legal or causal conclusion.
Fairness review belongs inside validation rather than after deployment.
What this chapter covers
- 01
Define job-relevant criteria before evaluating selection predictors
- 02
Link applicant, assessment, decision and outcome data with clear time order
- 03
Calculate within-group selection rates using the correct applicant denominator
- 04
Use categorical association and regression for questions their designs support
- 05
Check subgroup accuracy, missingness and stability in later cohorts
- 06
Investigate unequal patterns without inferring a mechanism from a rate alone
Worked example · free
Compare hiring rates with valid denominators
- +1The marks shown beside these steps are not a published assessment weighting. Group A selection rate = 36 divided by 120 = 0.30, or 30%.
- +1Group B selection rate = 16 divided by 80 = 0.20, or 20%.
- +1Percentage point gap = 30% minus 20% = 10 percentage points.
- +1Relative selection rate for Group B = 20% divided by 30% = 0.667, approximately 0.67.
- +1The observed table does not establish why the rates differ, whether measures are job relevant or whether a legal standard is met.
Key terms
- Selection rate
- The number receiving a selection outcome divided by the relevant applicant count within the same group and stage.
- Criterion validity
- Evidence that a selection measure relates to a relevant current or future performance criterion.
- Criterion deficiency
- Failure of the outcome measure to capture important parts of job performance.
- Criterion contamination
- Influence on the outcome measure from factors outside the performance construct of interest.
- Validation cohort
- A later or separate group used to test whether model performance generalises beyond development data.
Staffing and Selection Analytics FAQ
Why is later performance needed for selection validation?
A selection score is useful only if it supports the intended hiring decision. Linking it to a job-relevant later criterion tests that connection. The criterion must also be credible: biased supervisor ratings or training completion can make a model appear accurate while measuring opportunity or process compliance instead of performance.
Does an unequal selection rate prove discrimination?
No. It is an observed pattern that requires investigation. Review denominators, missing records, stage transitions, measure relevance, subgroup accuracy and how decision makers used the score. A statistical association alone does not identify the mechanism or settle a legal judgment.
Why test a model in a later cohort?
Development results can be optimistic and can depend on a particular applicant pool, job design or rating process. A later cohort tests stability under changed conditions. Monitor calibration, errors, missingness and subgroup performance, then document whether the measure still supports the original decision purpose.
What is incremental validity in selection?
Incremental validity asks whether a new assessment improves prediction or decision quality beyond information already available. The gain must be considered alongside cost, candidate burden and subgroup performance. A complicated measure is not valuable merely because its standalone association is statistically detectable; compare the full decision process with and without it.
Exam move
Sketch the selection pipeline from applicant pool to later performance and write the denominator beneath every transition. For each predictor, ask what job requirement it represents and what evidence supports that link. For each criterion, list one deficiency and one contamination risk. Practise calculating rates from two-by-two tables, then add a sentence separating the descriptive pattern from its possible explanations.
In exam answers, organise recommendations by stage: verify the source data, review the measure, inspect decision consistency, validate against later outcomes and monitor future cohorts. That structure turns a fairness concern into an auditable analysis plan. Build a full two-by-two table whenever a rate comparison appears, including selected and not selected counts.
Repeat the rates at screening, interview and offer stages to locate where a pattern begins. Then change the performance criterion and explain how the meaning of validation changes with it. Compare a predictor that is highly accurate against a contaminated rating with a modest predictor against a strong job criterion. Add a later cohort and inspect whether the result travels across time and groups.
Your final paragraph should state the observed disparity, the unresolved mechanism, the stage to audit and the safeguard that prevents an unvalidated score from becoming an automatic employment decision.
Working through Staffing and Selection Analytics in WORK6010? Sia is AskSia’s AI Management tutor — ask any WORK6010 Staffing and Selection Analytics question and get a clear, step-by-step explanation grounded in how WORK6010 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
