Check Population Margins Before Poststratifying a Survey

Match the population behind the margins

A population total is useful only if it describes the target represented by the survey. Check its reference date, geography, eligibility rules and category definitions. A count of all residents may not match a survey restricted to community-dwelling adults, even when both files use the same regional labels.

The survey poststratification documentation distinguishes matching a known joint distribution from matching separate marginal totals. When only margins are available, a raking procedure addresses a different information structure. Do not invent the missing joint cells by assuming that age, region and sex are independent.

Build a crosswalk before fitting weights

Create a table linking each survey category to its population counterpart. Check spelling, code values, age boundaries and reference periods. Identify population categories absent from the sample and sample categories absent from the target. These are analytical questions, not formatting defects to resolve by replacing missing counts with zero.

Verify that cell totals sum to the intended population. If several marginal tables describe the same target, check that their grand totals agree or document the reconciliation rule. Keep the original source tables alongside the transformed input so a reviewer can trace every category merge and total adjustment.

Reconcile a hypothetical two-cell example

Suppose the target contains 600 people in category A and 400 in category B. Before adjustment, survey weights sum to 500 in each category. Simple cell factors are 600/500 = 1.2 for A and 400/500 = 0.8 for B. The adjusted totals are 600 and 400, with grand total 1,000.

If category B has no sampled observations, multiplying its nonexistent observations by a large factor cannot create evidence. The team must reconsider the estimand, combine categories under a defensible rule, obtain more data or acknowledge that the intended calibration cannot be supported.

Review the output as an analysis object

After adjustment, compare achieved and requested totals with an explicit numerical tolerance. Inspect weight extremes, signs and any failed convergence. Confirm that the variance procedure recognizes the adjustment and that replicate weights, where supplied, receive compatible treatment.

Record whether margins came from a census, administrative system or another sample with its own uncertainty. Treating estimated control totals as fixed may omit that uncertainty. Report the target definition and unresolved coverage differences with the weighted results. Exact agreement with supplied margins demonstrates a computational constraint was met; it does not establish unbiased outcomes or eliminate nonresponse concerns.

Source verification: 6 November 2026. Examples are hypothetical.