Measured vs estimated impressions: how to report responsibly

A PSA campaign report holds two kinds of numbers. Some are observations: this spot aired on this station at this time. Others are calculations: this many people probably saw it. Both belong in the report, and broadcast impressions are almost always the second kind. Trouble starts when the two are presented as if they were the same thing.

This guide covers where impression figures come from, why they’re almost always estimates for broadcast PSAs, and how to present them so a board member, a funder or a journalist can tell what was observed and what was modeled. These are the rules PSAReach applies in its own reports. The methodology page sets them out formally.

What’s the difference between measured and estimated figures?

A figure is measured when it records something that happened and can be checked against a source. It’s estimated when it’s calculated from other data using assumptions, and a different but reasonable set of assumptions would give a different answer.

The line where measured rows give way to estimated ones is the most important line in the report. Anything below it needs a label, a method and a source period.

What is measured in a PSA report?

Airings are measured. A monitoring service records the broadcast, matches it to the creative and produces an occurrence record. Once that’s normalized, the airing count, the station list, the market list and the daypart breakdown are facts. You can quote them without qualification, apart from the monitoring coverage caveat, which a good report states once.

Conversions are measured when they come from the organization’s own accounts. A donation completed after someone clicks a search ad is recorded by the organization’s own website and its own Google Ads account. Nobody modeled it.

Digital reach and frequency are measured by the platform. When a display or video campaign runs through a major advertising platform, the platform reports how many unique people it reached and how often, based on its own logged-in user data. It isn’t perfect. It’s still observation rather than inference, so you can label it measured and name the platform as the source.

What is estimated, and why?

Nobody counts the viewers of a specific PSA airing. There’s no meter on the break. Audience research measures samples of households and people, projects them to the population and publishes ratings for programs and dayparts over a survey period. The impression figure for an airing is that published rating, or one derived from it, applied to the airing after the fact.

The estimate is softer than it looks, for three reasons.

• Ratings are sample-based. They carry sampling error, and the error is larger for small markets, small stations and low-audience time periods. That’s where donated airings tend to run.
• PSAs often air outside rated programs. Overnight and early-morning periods may have only a daypart average, with no program-level rating at all. The estimate then covers a broad block rather than the exact minute.
• The source period rarely matches the airing. A rating from one survey period gets applied to an airing that ran weeks or months later. Audiences shift, especially around seasonal programming and news events.

Impression estimates are still useful. The report just has to call them estimates.

How is an impression estimate built?

For television, the standard calculation multiplies a rating by a universe estimate. A rating is the percentage of the universe exposed to the average minute of a program or daypart. The universe estimate is the number of households or people in the market for the demographic in question.

Estimated Impressions per Airing = (Rating ÷ 100) × Universe Estimate
Gross Estimated Impressions      = Sum of Estimated Impressions across all airings

For radio, the equivalent input is the average quarter-hour audience, usually written AQH. It’s the average number of people listening to a station during any 15-minute period within a daypart. Each airing gets the AQH for its station and daypart.

Two decisions in this process should be visible in the report. The first is the grain. Was the airing matched to a program-level rating or to a daypart average? Program grain is more precise where it exists, while daypart grain is more widely available. A report that mixes them without saying so hides a real difference in confidence.

The second is the demographic. Household impressions and adults 25 to 54 impressions are different numbers, and the report should say which one it shows.

Why do estimates compound?

Impressions are the building block of the report. Nearly every other estimate starts from them. Estimated value is checked against impressions through CPM, estimated GRP is estimated impressions divided by the universe, and estimated reach is modeled from the schedule using assumptions about duplication. Each step piles its own uncertainty on top of the last.

That’s why a report should never show GRP or reach with more precision than the impressions underneath them. An impression estimate with a plausible error of plus or minus 20 percent can’t support a reach figure quoted to three significant figures. Round the derived figures.

How should an estimate be labeled?

A label is more than the word “estimated” in a footnote. A proper one has three parts.

1. The word itself. “Estimated impressions”, “estimated GRP”, “estimated value”, wherever the figure appears, including chart axes, table headers and exports.
2. The method. One line does it: “program-level rating times market universe” or “daypart AQH times airings”.
3. The source period. The survey period or data month the audience figures came from, and the date the rate evidence was compiled.

PSAReach reports carry this label on every estimated figure. Dashboards show an “estimated” badge next to the number instead of burying the caveat on a methodology tab.

What are the presentation rules for estimates?

Four rules head off most of the misuse that happens once a report leaves the analyst’s hands.

• GRP can be summed. Reach can’t. GRP is a gross measure, so you can add it across weeks, stations and markets. Reach counts unique people. Add this week’s reach to last week’s in the same market and you’ve counted the same people twice.
• Reach adds across markets but not across time. Each county belongs to exactly one DMA, so the people in two different markets are different people. Summing market reach into a national reach figure is sound arithmetic. Summing weekly reach within one market isn’t.
• Spot length changes value but not audience. A :60 in the same break as a :30 reaches the same people. It’s worth more, but it delivers the same estimated impressions. Never scale impressions by length.
• Don’t combine cross-media reach. Without single-source data, there’s no way to measure the overlap between people who saw a TV PSA, heard a radio PSA and saw a display ad. Report each channel’s reach separately rather than inventing a combined figure.

A worked example

Suppose a :30 PSA airs in a 6 a.m. newscast in a mid-size DMA. The most recent survey period gives that newscast a 2.1 household rating, and the market’s universe estimate is 640,000 television households.

Estimated Impressions per Airing = (2.1 ÷ 100) × 640,000 = 13,440 households

Over the flight, the spot airs 25 times in that newscast.

Gross Estimated Impressions = 25 × 13,440 = 336,000 household impressions
Estimated GRP               = (336,000 ÷ 640,000) × 100 = 52.5

The report line reads: “25 airings (measured) delivered an estimated 336,000 household impressions and an estimated 52.5 GRP, based on the program-level household rating from the most recent survey period.”

Reach needs a model. Twenty-five airings in the same newscast reach many of the same households again and again, so the unique audience is far smaller than 336,000. A duplication model that reflects that concentration might put estimated reach at about 26 percent of households, roughly 166,000, with an average frequency of about two. Report it as an estimate. Quoting “166,000 people reached” with no “estimated” beside it claims more than anyone knows.

If the same campaign ran in a second DMA, its estimated reach there could be added to this one, because the households are different. If it ran a second flight in this DMA a month later, the two reach figures couldn’t be added, because most of the households are the same.

A checklist before the report goes out

Before a PSA report goes to leadership or a funder, check these points.

1. Each airing figure traces back to measured occurrence data, and the monitoring coverage caveat is stated once.
2. Each impression, value, GRP, TRP, reach and frequency figure carries the word “estimated” wherever it appears.
3. The method and source period for audience estimates are stated.
4. The grain (program or daypart) and the demographic (households or a target) are stated.
5. Value is labeled estimated, with its rate basis named.
6. Reach is reported per market and per period, never summed across weeks.
7. No combined cross-media reach figure appears.
8. Derived figures are rounded to a precision the inputs can support.
9. Measured conversions from the organization’s own accounts sit in their own section.
10. Exports and charts carry the same labels as the narrative.

A report built this way takes longer to write, and it holds up far better when someone starts asking questions. For the full set of definitions, see the glossary. For how PSAReach produces the audience and rate evidence behind these estimates, see Media Intelligence.