The Local Base Beneath the National Figure: Auditing Regional Evidence Before a UK Rollout

The board pack arrives with a familiar page: a UK map shaded from pale to dark, a national sample of 2,000 adults, and a recommendation to launch in three regions. The national figure is defensible. The regional picture may not be. A UK-wide survey can support a national estimate while leaving individual regions with bases too small to carry a launch decision. The audit below is designed for that moment — before a regional rollout is approved, not after the map has become a commitment.

Begin with the base behind the colour

Every regional estimate rests on a subgroup base. In a typical UK-wide online survey of 2,000 respondents, a Government Office Region may hold 150 to 250 interviews; a city or combined authority may hold fewer still. At a 95% confidence level, a 50% estimate from a base of 150 carries a margin of error of roughly eight percentage points. At 100, it is closer to ten. These are not exotic statistical points. They are the difference between a region that clearly clears a launch threshold and one that may or may not.

Weighting complicates the picture. A sample weighted to age, sex, region and education may be more representative, but weighting can reduce the effective base and widen intervals. The board should ask for both unweighted and weighted bases, and for the effective base where weighting has been applied. A regional sample of 180 that weights down to an effective base of 110 is not the same evidential asset as a straightforward sample of 180.

A quota is not a guarantee of local coverage

Regional quotas are often set at a broad level. A survey can fill a North West quota without representing the difference between Greater Manchester and rural Cumbria. For a city launch, the relevant geography may be a local authority, a combined authority or a set of postcode districts. A regional quota does not guarantee interviews in those places.

Before the board treats a regional result as local evidence, ask how the quota frame was built. Were quotas set by region only, or also by postcode sector, urban and rural mix, or local authority? Were achieved interviews reported against those quotas? If the rollout footprint is a handful of cities, a regional average is a weak substitute for city-level bases. The map may look precise because it has boundaries. The evidence may be much coarser than the boundaries suggest.

Confidence intervals at the decision threshold

Regional launch decisions usually have a threshold: awareness, trial, availability or willingness to pay. The board should see an interval around the regional estimate and compare that interval with the threshold. If the threshold is 40% and the estimate is 44% with an interval of 36% to 52%, the evidence is compatible with both launch and no launch. A point estimate alone invites false confidence.

This is why the threshold should be agreed before the results are seen. If the board sets the bar after viewing the map, the interval becomes a debating prop rather than a decision rule. A short note in the board pack can state the threshold, the regional estimate, the interval and the base. That is not statistical decoration. It is the minimum required to distinguish a measured regional signal from a national average redistributed across a map.

Separating evidence from interpolation

Many maps interpolate. A choropleth may colour every local authority in a region from a regional estimate. That is a presentation choice, not a measurement. Ask the insight team to mark areas with direct evidence, areas with modelled estimates, and areas with no usable base. Use hatching or a neutral fill for the third category. Label bases on the map or in a companion table.

A useful discipline is to produce two pages: one map for communication and one base table for audit. The map can show the pattern the board needs to discuss. The table should show the unweighted base, weighted base and interval for each area that carries a decision. If an area has a base below 100, the map should say so. A colour that implies local precision from a thin base is not a neutral design choice; it is an evidential claim.

A short audit routine for the board

1. Request unweighted and weighted bases for every region, city and postcode district that will appear in the decision.
2. Ask for 95% confidence intervals at the level of the decision, not only at national level.
3. Check quota frames against the rollout footprint. A region quota is not a city quota.
4. Identify any estimate that is modelled, interpolated or borrowed from a neighbouring area.
5. Compare the interval with the pre-agreed launch threshold.
6. Decide whether a regional booster is cheaper than the risk of a wrong launch.

When to commission a booster

If the rollout commitment is significant, a regional booster is often a small addition to the research budget. But a booster should be designed for the areas that will carry the decision, with quotas at the right geography and fieldwork dates that match the launch window. Adding interviews to an existing study is not automatically sufficient if the original sampling frame did not reach the target postcodes.

The board does not need to become a statistics seminar. It needs a clear separation between what was measured, what was modelled and what remains unknown. The map is a communication device. The base table is the evidence. A board that asks for both — and treats the difference as material — can approve a regional rollout with a clearer view of what is known, what is inferred and what remains a gamble.