WORK6010 Chap.2 Data Collection, Sampling and Measurement
Data Collection, Sampling and Measurement
HR records are not neutral snapshots of the workforce. They are produced by sampling choices, administrative processes, survey wording, timing and access rules. This chapter links reliability and validity to the practical decisions that create data.
It shows why a large dataset can still be biased, why a convenient proxy can miss the intended construct and why missingness should be investigated before rows are removed or imputed.
What this chapter covers
- 01
Define the target population and the mechanism that places cases in the sample
- 02
Distinguish a construct from the operational measure used to represent it
- 03
Compare reliability, content validity, construct validity and criterion validity
- 04
Diagnose coverage, nonresponse, recording and timing bias in HR data
- 05
Describe missing data patterns before choosing deletion or imputation
- 06
Use data minimisation and access controls as part of measurement design
Worked example · free
Diagnose a biased engagement pulse
- +1The marks shown for this audit are not a published University weighting. Response rate = 180 divided by 600 = 0.30, or 30%.
- +1Observed high-engagement proportion among respondents = 135 divided by 180 = 0.75, or 75%.
- +1Keep the denominator explicit: 75% describes respondents, not all 600 employees.
- +1Identify possible nonresponse bias because the likelihood of responding may be related to engagement, device access or working conditions.
- +1Compare respondent and workforce profiles, improve access and follow up nonrespondents before generalising to the whole workforce.
Key terms
- Target population
- The full set of people or records to which the intended conclusion applies.
- Sampling frame
- The operational list or process from which observed cases can be selected.
- Measurement reliability
- The consistency of a measure across items, raters, occasions or equivalent procedures.
- Construct validity
- Evidence that the operational measure represents the intended underlying concept.
- Nonresponse bias
- Distortion that occurs when response is related to the outcome or relevant characteristics.
Data Collection, Sampling and Measurement FAQ
Can a very large HR dataset still be biased?
Yes. Size reduces some sampling variability but does not repair systematic exclusion, selective response or inconsistent recording. A payroll file can be nearly complete yet omit contractors; an optional survey can be large yet overrepresent office staff. Representativeness depends on the process that generated the data.
What is the difference between reliability and validity?
Reliability concerns consistency. Validity concerns whether the measure supports the intended interpretation and use. A performance rating can be highly consistent if every manager applies the same narrow rule, yet invalid for overall job performance because the rule omits important responsibilities.
Should missing rows always be deleted?
No. First describe how much is missing, which variables and groups are affected, and why values may be absent. Complete-case deletion can change the analysed population and bias results. The response should match the missingness process, decision stakes and available information, with assumptions reported clearly.
What should an HR data audit document?
Document the target population, collection channel, operational definitions, exclusions, missing-value codes, time window, aggregation and access. Follow a record from entry to report and note where meaning can change. This creates a reproducible explanation of the dataset and directs repairs toward the process that created bias or ambiguity.
Exam move
Draw a measurement target for each applied topic. Put the construct in the centre, then surround it with the actual fields used, the people excluded, the timing and the recording process. Mark each threat as coverage, nonresponse, measurement or missingness. Next, practise converting percentages back to counts and denominators so you can see what a summary really represents.
In revision answers, use one sentence to define the target population and another to describe how the observed sample arose. Only then discuss a test or model. That order earns analytical credibility because it shows whether the numbers can answer the business question at all. Build one audit trail from an imagined employee response to the final dashboard.
Note the invitation channel, response option, validation rule, missing-value code, aggregation and access decision at every step. Change one feature, such as moving a survey to mobile-only delivery, and predict who becomes less observable. For administrative data, compare the policy definition with the system field and list cases that fall between them.
Practise answering with counts as well as percentages because 30% from 18 people and 30% from 1,800 people have different precision and disclosure risks. Close the exercise by specifying a repair that targets the data-generating process, not a statistical decoration applied after collection.
Working through Data Collection, Sampling and Measurement in WORK6010? Sia is AskSia’s AI Management tutor — ask any WORK6010 Data Collection, Sampling and Measurement 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.
