On this page
Describe how selection actually occurs
Record the randomization rule and the frame to which it applies. A reproducible design specifies what could have been selected, not merely which units were selected this time. The probabilities must concern the units in the defined survey population.
Do not require equal probabilities for every design
Probability sampling can use unequal known positive inclusion probabilities. For example, a design may select more units from a small subgroup to study that subgroup. Estimation then needs the design information rather than assuming the collected rows represent population shares.
Name the quantity and its estimator
An unweighted sample mean, a population-weighted stratum mean and a design-weighted total are different calculations. Choose the estimator to match the target and selection rule. A table of sampled values without selection information can conceal that distinction.
| Item | Question |
|---|---|
| Population quantity | Mean, total or subgroup contrast? |
| Selection rule | Which sample sets can occur? |
| Probabilities | What is known for each unit? |
| Estimator | How do selected units contribute? |
| Uncertainty | Which design is assumed? |
Keep a random design separate from a realized sample
A valid random design does not guarantee that every realized sample looks like the population in every feature. Report the design and any observed imbalances. Do not substitute a reassuring description of a sample for the actual probability mechanism.