Allocate the sample for the subgroup decisions the study must support

Translate each reporting promise into an allocation need

Write the decisions the research must support before choosing a total sample size. A broad satisfaction description, a comparison between service tiers and a detailed account of a small recently launched tier demand different evidence. Identify which outputs are essential and which are exploratory. A promise to report every combination of customer characteristics can create more thin cells than the budget can meaningfully support.

Ask the analyst to explain the precision or resolution needed for each essential output in practical units. Record assumptions about outcome variation and expected response, identifying estimates rather than treating them as known facts. Keep selected sample size distinct from completed responses. A plan that selects enough units under an optimistic response assumption needs a contingency, not a later claim that whatever arrived must be sufficient.

Define strata available before selection

SAS/STAT 14.1 documents independent sample selection within nonoverlapping strata and distinguishes proportional allocation from alternatives using stratum variability and cost. These are versioned methodological illustrations, not instructions to apply a software default.

Choose permitted frame variables that can be assigned consistently before the draw. Define boundaries, missing classifications and the reference date. Every eligible sampling unit needs one unambiguous location in the design. Do not move accounts between strata because of answers received later. If a frame field is unreliable, repair or simplify the planned grouping with the owner rather than concealing uncertainty in a miscellaneous category.

Distinguish a design stratum from a reporting domain that will be identified only through questionnaire answers. The latter may matter greatly, but its eventual count cannot be promised by pretending it was known at selection. Preserve this distinction in the allocation brief so a stakeholder does not mistake a desired audience breakdown for an operationally achievable sampling guarantee.

Compare allocations against the whole decision portfolio

Lay out population counts, planned selected counts and expected achieved counts separately for each stratum. Ask the analyst to compare a population-proportional allocation with a justified alternative that gives a small but important stratum more observations. Evaluate the trade-off using the intended outputs and fieldwork costs, not whether the sample visually resembles the customer base. An intentionally disproportionate sample may be useful, provided its selection structure survives analysis.

Check feasibility with the recruitment owner. A small stratum cannot supply unlimited eligible distinct units, and selecting most of it does not guarantee that most will respond. Record capacity limits, permitted contact arrangements and the approval needed if recruitment falls short. If the budget cannot support every promised comparison, narrow the reporting commitment before fieldwork rather than hiding the shortage behind weighted percentages afterward.

Freeze the allocation and hand over its consequences

Approve a versioned allocation table with stratum definitions, selected counts, within-stratum selection method and the analyst’s treatment of unequal inclusion chances. Preserve the actual draw and deviations alongside the plan. Fieldwork staff should not transfer unused places from a difficult stratum to an easier one without a documented research decision; the resulting sample would answer a different allocation plan.

At handover, reconcile planned selections, actual selections and completed interviews within each stratum. Explain where a domain output must be withheld or qualified. Do not interpret oversampling as proof that a subgroup differs, or weights as evidence that every missing experience has been recovered. The useful artifact makes the allocation’s purpose and cost explicit while giving the analyst the information needed to represent the population honestly.

Make the reasoning concrete.

A fictional subscription service wants an overall onboarding readout and a separate decision about a small enterprise tier. A proportional allocation would leave few selected enterprise accounts.

The team chooses a justified allocation and revises unsupported reporting promises before collecting answers. It retains selected and achieved counts separately.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Outputs

   Which subgroup decisions must the study actually resolve?

2. Strata

   Are group definitions exclusive and available before selection?

3. Counts

   What are population, selected and expected completed counts?

4. Trade-off

   Which outputs or costs change under the proposed allocation?

5. Analysis

   What selection fractions and deviations must accompany the responses?

Watch for the wrong conclusion.

• Treating an achieved quota as a documented probability draw.
• Promising a precise small-group result from an overall sample target.
• Moving selections between strata without recording a design change.

Sources & scope

Workflows, worksheets and fictional examples are original editorial analysis. Sources support only specifically attributed principles. Platform guidance can change; legal requirements depend on jurisdiction and context. No measured results or named human reviewer are claimed.

• SAS/STAT 14.1 User’s Guide: STRATA Statement →Primary methodology documentation · Supports: Versioned independent stratified selection and allocation distinctions.