SO5005 Chap.5 Sampling and Sample Size
Sampling and Sample Size
Track three populations
The target population is named by the claim, the sampling frame is who can be reached, and the achieved sample is who contributes usable data. Gaps between them create coverage, non-response and attrition problems.
Probability sampling concerns a known selection mechanism, not merely a large number of observations.
Plan size for the primary analysis
A calculation needs the primary outcome, meaningful effect or precision, variability or event rate, significance level, power and an allowance for loss. Inflating recruitment protects quantity, but biased loss still threatens validity.
What this chapter covers
- 01
Target population
- 02
Sampling frames
- 03
Probability sampling
- 04
Purposeful and convenience routes
- 05
Sample-size assumptions
- 06
Attrition and retention
Worked example · free
Inflate a recruitment target
- 2Convert loss to retention.
- 2Divide the complete target by retention.
- 2Interpret what the calculation cannot fix.
Key terms
- Target Population
- The complete group to which a research conclusion is intended to refer.
- Sampling Frame
- The operational source from which eligible units can be selected or contacted.
- Attrition
- Loss of enrolled participants or observations before the planned analysis is complete.
Sampling and Sample Size FAQ
Does a larger sample remove sampling bias?
A larger sample reduces random uncertainty when the design and model are suitable, but it does not automatically improve coverage or response. A huge convenience sample can estimate its respondents precisely while representing the target population poorly.
Why use stratified sampling?
Stratification separates the frame into important subgroups and samples within each one. It can secure adequate representation for comparisons and improve precision when strata are informative. Whole-population estimates may require weights that reflect the sampling fractions.
What assumptions belong in a sample-size section?
State the primary outcome and analysis, target effect or precision, variability or event rate, significance level, power and expected unusable data or loss. Each assumption needs a defensible source, and the final recruitment number should be distinguished from the analysed target.
Assessment move
Draw the target population, frame and achieved sample for a proposed study. Mark who disappears at each transition. Then recompute one recruitment inflation and explain separately what random uncertainty, coverage and attrition each calculation or design step addresses. Start with the exact noun phrase used in the intended conclusion and place it above the target-population box.
List every route by which an eligible person could enter the frame. Missing routes reveal coverage problems before recruitment begins. Next, simulate a simple random sample with numbered slips, a systematic sample with a random start, and a stratified sample with different sampling fractions.
Explain which procedure gives a known selection probability and why convenience remains convenience even when the achieved number is large. For cluster sampling, identify the natural groups and state why people within a group may resemble one another. Keep sample-size practice tied to a primary outcome.
Write the required analysed number, expected effect or precision, variability or event rate, significance level and power on separate lines. Treat each as an assumption that could be challenged. For loss, convert the percentage to retention and divide the complete target by that fraction, then round upward. Check by multiplying the recruitment number by retention.
Distinguish this quantity adjustment from attrition bias: more recruits cannot restore representativeness when loss depends on outcome or tolerability. Plan a response table that accounts for eligible, approached, consented, measured and analysed people. Differences between these counts belong in the result, not only in project administration.
End by writing one paragraph that tells a reader who the final estimate represents and one sentence naming the group it may fail to represent. That boundary is part of the finding. Create alternative sampling plans for the same question under limited time, restricted access and a need for subgroup comparison. State which source of error each compromise increases.
For a probability design, practise calculating selection fractions and explaining whether weights are needed. For a non-probability design, avoid probability language and justify why the recruited cases remain informative. Draft recruitment materials and compare their promised benefits with what the study can genuinely offer.
Add a contingency plan for slower response that preserves eligibility and selection rules rather than replacing difficult participants with convenient ones. Before analysis, reconcile the recruitment log with the dataset so exclusions are traceable. This makes the final sample description a reproducible result instead of an administrative afterthought.
Keep the flow diagram and denominator table together so a reader can reconcile every count from initial eligibility to the analysis that produced the reported estimate. Differences must be explained rather than left as arithmetic puzzles. Each transition needs a reason. Keep every count reproducible from the recruitment log.
Working through Sampling and Sample Size in SO5005? Sia is AskSia’s AI Statistics tutor — ask any SO5005 Sampling and Sample Size question and get a clear, step-by-step explanation grounded in how SO5005 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
