Marketing mix modelling England Scotland Wales: a guide

What to take away

Why devolved media channels change MMM inputs

Devolved media channels are not a marketing curiosity. They change what you can put into a model and what the model can tell you. Health, education and much cultural advertising in Scotland and Wales runs through bodies that answer to the Scottish Government or the Welsh Government, not to Westminster.

Their campaigns have different timing, different audience definitions and different reporting.

That matters for marketing mix modelling. A UK brand running one campaign across three nations may find that the Scottish leg starts three weeks earlier because a public health body there set the schedule. If your model uses a single national start date, it will attribute the early lift to the wrong channel.

The constitutional basis for that difference is the Scotland Act 1998, which created the Scottish Parliament and its devolved competences. The equivalent for Wales is the Government of Wales Act 2006, which gave the Senedd and Welsh Ministers their current legislative framework.

Both Acts are why a planner cannot assume that a UK government communication plan covers all three nations.

In practice, you should map every input to the nation or nations it actually covers. A media plan that says "UK" is not an input. It is an assumption you have not tested.

England needs explicit geographic coding too, rather than being treated only as a residual. Map campaigns commissioned by NHS England, the Department for Education or Arts Council England to their actual English coverage, and keep UK-wide activity separate.

The same logic applies to owned and earned channels. A retailer with separate Scottish and Welsh store estates may run separate price promotions. If those promotions do not appear in the model, their effect lands on whichever media variable happens to move at the same time.

For panel-based sales data, ask NielsenIQ and Circana whether the specific UK product and category can be reported separately for England, Scotland and Wales. Confirm the coverage, dates and definitions with the supplier rather than assuming a panel supports nation-level reporting.

For a wider grounding in the method before you localise it, see our guide to marketing mix modelling.

STV and S4C as separate media variables

STV and S4C are the clearest examples of devolved media channels that deserve their own variables. STV holds the Channel 3 licence for central and northern Scotland and sells its own airtime. S4C is the Welsh-language public service broadcaster, funded by licence fee and advertising, with its own schedule and audience.

If you fold either into a generic "TV" line, you lose the ability to see whether it worked. Worse, you can create a false result.

Suppose a brand buys STV heavily in a quarter when Scottish sales also rise because of a local distribution deal. A combined TV variable will absorb that lift and make television look more effective than it was.

Build the inputs separately:

STV advertising measurement also has a practical wrinkle. Its reach in Scotland is high, but its absolute impressions are small next to a UK-wide channel. That means noisy weekly data. Weekly granularity may be too fine. Consider fortnightly or monthly aggregation for the STV variable, and say so in your documentation.

S4C campaign measurement has a different wrinkle. Its audience is concentrated among Welsh speakers and learners, so a national Welsh sales series is too broad a dependent variable. If you can get region-level or language-segment sales, use them. If you cannot, treat the S4C coefficient as directional only.

Both broadcasters sit inside a wider set of devolved media channels, including Scottish and Welsh news titles, community radio and out-of-home estates that differ from England. Each one you can separate is a chance to learn. Each one you merge is a chance to be wrong.

Name the relevant devolved services in the plan: BBC Scotland and BBC Alba, alongside STV, in Scotland; BBC One Wales, BBC Radio Wales and BBC Radio Cymru, alongside S4C, in Wales. For audience data, BARB measures television, RAJAR radio, Route out-of-home (OOH) and JICMAIL direct mail. Match each measure to the geography and period supplied; do not assume every source provides a separate series for each nation.

Before you sign any contract for the audience data behind these variables, our notes on effectiveness resources set out what to verify.

Regional price differences and deflator choices

Regional price differences are the quiet killer of three-nation models. You are usually modelling sales value, but media and consumer behaviour respond to real volumes and relative prices. If a Glasgow household and a Bristol household both spend £40, they have not bought the same basket.

A single UK deflator will therefore distort your dependent variable. In periods when Scottish or Welsh prices move differently from the England average, the model reads a price effect as a media effect. The fix is to deflate each nation's sales series with a nation-appropriate index before modelling, and to keep the deflator documented.

Three practical choices:

The third option is often the most defensible because it matches your actual product mix. It also keeps you honest: if your Scottish average selling price diverges, you can see it rather than assume it away.

Media costs need the same treatment. STV airtime and S4C airtime are priced in the same currency as English airtime but not in the same market. A cost-per-thousand comparison across nations without a price adjustment will flatter whichever nation had the cheaper inventory that quarter.

Write the deflator choice into the model register. State the source, the base period and the nations covered. When someone reruns the model next year, they should not have to guess.

Population and GVA denominators across the three nations

Denominators decide what "good" means. If you measure sales per head, England, Scotland and Wales have different population bases and different age structures. If you measure sales per unit of economic activity, you need Gross Value Added.

ONS GVA data is the standard starting point for regional and devolved-nation benchmarking. It lets you express performance relative to the size of the local economy rather than raw totals. For city and region comparisons across England, Scotland and Wales, the ONS regional accounts are the companion dataset.

Use them like this:

GVA regional benchmarking also exposes a trap. A nation with lower GVA per head is not automatically a weaker market. It may have lower costs, different retail density or a younger population. The denominator is a lens, not a verdict.

For Welsh-specific campaign measurement, the Welsh Government statistics and research often run on different definitions and release cycles from ONS products. Where a devolved statistic exists, prefer it for devolved campaigns and keep the ONS series for cross-nation comparison.

For Scotland, check Scottish Government statistics for devolved campaign measures; for England, use the relevant English department or body where its data is specific to the campaign. Keep ONS series as the consistent cross-nation comparison where definitions allow.

One more point on population. Mid-year estimates for the three nations are revised, and internal migration has shifted patterns in recent years. If your model uses a fixed population weight for the whole window, a revision can move your per-head results without any change in sales.

For population denominators, name the ONS mid-year population estimates in the register and record the release used, so revisions can be traced.

Reading MMM outputs when one nation is a small share

Scotland and Wales are small shares of most UK brands' sales. That creates two problems: weak signal and unstable coefficients.

Weak signal means the model may not detect a real effect. A channel that genuinely lifts Welsh sales by a few per cent can be lost in noise if Wales is 4 per cent of volume.

Unstable coefficients mean the model may detect an effect that is not there, because a small series can be fitted by many combinations of variables.

Four ways to handle it:

A worked example makes the risk concrete. Suppose total UK sales are 1,000,000 units. Scotland is 90,000 and Wales is 45,000. A campaign runs on STV and S4C only. The model reports a strong English media coefficient because English media also moved that month, and near-zero coefficients for STV and S4C.

The planner concludes the devolved channels failed. In truth, the model had too little Scottish and Welsh variation to separate them from the English trend.

The remedy is not to abandon the question. It is to design the test so the devolved channels move independently, for example by running a holdout in one nation while the other continues. That gives the model the contrast it needs.

Keep an internal baseline for each nation so you can judge whether a reported lift is larger than normal week-to-week movement. Without it, every small-nation result looks like a finding.

A three-nation input register

A register is the simplest way to stop inputs drifting between nations. It lists every variable, its source, its geography, its deflator and its owner. Build it before you build the model.

Use the register as a checklist before each model refresh:

Two governance points sit above the register. First, UK GDPR and the Data Protection Act 2018 apply across all three nations, but the Information Commissioner's Office regulates them as one UK regime. Do not assume a separate Scottish or Welsh data regulator.

Second, the Competition and Markets Authority and the Digital Regulation Cooperation Forum shape how audience and ad data can be shared, and their guidance applies regardless of nation.

Finally, the Advertising Standards Authority, the Institute of Practitioners in Advertising and the Market Research Society set standards that your measurement partners should meet. A Scottish or Welsh campaign is not exempt from them because it is smaller.

Common questions