Verify the size measure that gives a sampling unit more chances

Identify the size concept and the sampling unit

Ask the analyst why the proposed size measure is relevant to this design. Define its unit, period and coverage in plain language. Active subscriptions, annual transactions and occupied workstations describe different aspects of an organization. Do not substitute a convenient count without design approval.

SAS/STAT 14.1 distinguishes the PPS size measure from the number of units selected and documents consequential handling of missing or nonpositive size inputs. These versioned facts reinforce the need to review input meaning; they are not universal preprocessing instructions.

Match the measure to the actual sampling unit. A parent company total copied onto each branch may make every branch appear as large as the company. Conversely, a branch-only count attached to a parent entity may omit much of its relevant activity. Retain the mapping evidence and have the source owner resolve ambiguous organizational boundaries before treating the values as interchangeable.

Check the reference date and construction history

Record when the size measure was observed, which processing steps produced it and which frame version it accompanies. Inspect whether new units, closures, mergers or transfers occurred between measurement and selection. A recent export does not mean every underlying value is recent. Track records whose apparent currency differs from their observation date.

Ask whether values share the same counting rules. One system might count all registered users while another counts only users active during a stated period. Preserve known differences rather than silently multiplying or imputing values to make the column look complete. If harmonization is justified, record the transformation, its assumptions and the owner who accepted it. Keep original source values available through the approved controlled workflow.

Resolve consequential exceptions before the draw

Flag missing, zero, negative and unusually large measures for investigation. A zero may describe a genuinely inactive unit, an unmeasured new account or a failed join. Those are different circumstances with different design consequences. Do not automatically exclude every zero or assign a convenient positive constant. The survey specialist needs the verified meaning before deciding how the unit remains eligible and how its selection chance is handled.

Trace an unusually large value to permitted source evidence. Check duplication, parent-child mismatches, unit conversion and genuine scale. A real large unit is not an error simply because it would be selected more often under the proposed method. If the design has special treatment for very large units, require that treatment to be documented by the analyst rather than inferred by an operations reviewer. Record unresolved meaning before the draw.

Freeze approved inputs and retain their analytical meaning

Produce a draw-ready acceptance register showing which measures were accepted, corrected, held or handled under an explicit design rule. Keep the approved selection-input snapshot distinct from later business updates. After selection, replacing the original measure with a current value can erase the basis on which the sample was chosen. Preserve selection evidence separately from later measurements.

Have the analyst confirm that the final input meanings match the planned design and that unresolved cases do not invalidate the intended use. If a material input error is discovered after a draw, pause the affected release and obtain a documented remedy; do not silently rerun until the selected list looks plausible. The practical outcome is trustworthy input lineage and a bounded release decision, not proof that PPS is always preferable or that the final sample will reflect every customer experience.

Make the reasoning concrete.

A fictional business-service survey plans to select locations using active subscription counts. One exceptionally large location carries its parent company’s total, while several new locations have blank counts.

The team chooses the exception review and preserves the original and accepted input versions. It does not claim that clean numeric values alone establish a valid design.

Take the questions into your next brief.

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

1. Concept

   What exactly does the size measure count?

2. Unit

   Does each value belong to the sampling unit rather than its parent or child?

3. Time

   When was the underlying measure observed?

4. Exceptions

   What do blank, zero and unusually large values mean?

5. Release

   Which verified snapshot and specialist decisions support the draw?

Watch for the wrong conclusion.

• Treating a current export date as a current measurement date.
• Giving all missing values an arbitrary positive size.
• Replacing selection inputs with later values and losing the original basis.

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: SIZE Statement →Primary methodology documentation · Supports: Versioned PPS size-input meaning and missing/nonpositive-input consequence; not universal preprocessing.