Reading a benchmark study sample definition before you quote the number

The slide says the industry expects rented space to grow by five per cent, so the show's target is five per cent. Somebody asks where the figure came from and gets a report name. Nobody asks the next question, which is who answered it.

A benchmark study sample definition is the part of a published report that tells you what its average is an average of, and in the better studies it is printed in full. Reading it takes ten minutes and it regularly changes what you would do with the number.

The four questions worth asking of any sample

Before a published figure goes into a target, a board pack or an exhibitor conversation, get answers to four things.

• How many usable responses. Not the invitation list, not the panel size. The count the percentages were computed on.
• Which markets they came from, and in what proportion. A global average built mostly from one region is a regional average with a global label.
• How respondents were recruited, and whether they chose to take part. A survey circulated through trade association membership reaches association members, who are systematically larger and better resourced than the market as a whole.
• What the publisher itself says about representativeness. Good publishers write the caveat down and most readers skip it.

The response rate, meaning responses against invitations, is a fifth question and a large enough subject on its own that P21 takes it.

A sample you can actually see

The 37th UFI Global Exhibition Barometer, published in July 2026, is a useful worked case because it publishes almost everything on the list.

UFI states that the study is based on a global survey concluded in June 2026 and reports on the global industry as well as 20 specific markets and regions. It records 466 companies participating from 59 countries and regions, and its appendix prints the number of replies per country.

Work through that appendix and the shape of the sample becomes concrete. Europe supplies 178 of the 466 responses, which is 38.2 per cent. Asia-Pacific supplies 113, or 24.2 per cent. North America supplies 62, Central and South America 57, and the Middle East and Africa 56. Inside those regions the distribution is uneven in ways that matter to a reader in a specific market: Turkiye alone accounts for 45 replies, Italy and China 29 each, Germany 20, Spain 19, and the United States 18.

That last figure is the one worth sitting with. The United States contributes 18 of 466 responses, which is 3.9 per cent of the sample, and 2.5 times fewer than Turkiye. A United States organiser reading a global figure from this study is reading a number that their own market contributed less than a twenty fifth of. The study is not misleading anybody about this, because it prints the table.

The report also names the associations whose participation made the collaboration possible, which tells you the recruitment channel: national and regional industry associations circulating the survey to their members. That is a legitimate method and it defines the population that could have answered.

One more property of a repeated survey is easy to miss. Each wave draws a fresh sample, so a movement between waves mixes a change in the market with a change in who replied. This edition has 26 per cent of respondents globally expecting activity in their home market to rise by more than 5 per cent in 2026, against 44 per cent six months earlier, and 21 per cent expecting a fall of more than 5 per cent against 8 per cent in the previous wave. Those are large moves and the report attributes them to conditions in the Middle East, which is plausible. Some part of an 18 point swing in any repeated survey is also composition, and no reader outside the publisher can separate the two.

How much does one respondent move the number?

This is the arithmetic that turns a sample size from a number into an intuition.

In the full sample of 466, one respondent changing their answer moves a percentage by 1 divided by 466, which is 0.21 points. Rounding alone absorbs it, and the global figures are stable in the sense that no individual company's mood is visible in them.

In the United States sub-sample of 18, one respondent moves a percentage by 1 divided by 18, which is 5.6 points. Three respondents changing their minds moves it by 16.7 points. Any market level figure computed on 18 responses can shift by more in a single wave than most real market changes do in a year, and the shift will look exactly like a trend.

Run the same calculation on whichever market you care about before you quote a market level figure from any study. Where the per country counts are not published, that calculation is unavailable to you, which is itself the answer about how much weight to give the number.

Two sample types, two different biases

Not every benchmark is a survey, and the second kind fails differently.

UFI's Euro Fair Statistics, in the 2022 edition published in November 2023, contains the certified statistics of 1,826 exhibitions from 14 countries. Every unit in that dataset has an audit behind it, so nobody is reporting their own opinion of how the year went. The sample is defined by which countries have a national reporting body feeding data in.

Compare the two designs honestly. The Barometer has broad geographic reach, 59 countries and regions, and depends on self reported perception from companies that chose to reply. Euro Fair Statistics has narrow geographic reach, 14 countries, and rests on audited figures for individual exhibitions. Neither is the better sample in general. The first tells you about sentiment across many markets, the second tells you about measured outcomes across a few, and quoting either as though it were the other is where most misuse starts.

The practical rule is to match the sample type to the claim. A statement about what organisers expect needs a survey. A statement about how much space was actually sold needs certified figures.

What does the publisher already admit?

The best evidence about a sample's limits usually comes from the publisher, printed in a section nobody quotes.

UFI writes of the 37th Barometer that "the nature of the exercise, a survey towards a broad sample of companies from the industry, means that some results cannot claim to necessarily be fully representative" (UFI, 2026). It goes further in the same passage, noting that "since weighting of answers based on the size of the responding company or its national market is not possible, it shall be noted that some regional or global consolidations, especially quantitative results, are not as accurate as the ones for the detailed markets" (UFI, 2026).

Read that carefully and it contains a specific instruction: prefer the detailed market results over the global consolidation. Almost every deck I have seen quoting this study does the opposite, because the global figure is a single number and the market tables take a paragraph to explain. The weighting question underneath it is P22's and it is worth reading before you compare any average against your own show.

A publisher who prints a caveat like that has done their job. A study with no methodology section at all is the one to worry about, and the correct treatment of a figure with no visible sample is to leave it out.

Quote the sample in the same sentence as the number

The habit that fixes most of this costs one clause.

Write "466 companies across 59 countries and regions, surveyed in June 2026, split 26 per cent expecting growth above 5 per cent" rather than "the industry expects growth". The sample travels with the figure into the next deck, the next email and the next board pack, and whoever reads it there gets the same ten second calibration you had.

The same habit applies internally. When your own analysis quotes a benchmark, put the sample in the footnote of the slide and in the comment of the query. A number without its sample will be reused; a number with its sample gets questioned, which is the outcome you want.

Where your show has no business being compared against the sample at all, the honest move is to stop and build your own baseline instead, which is what P20 argues.

Where this stops

Reading the sample tells you how much confidence a figure deserves. It does not tell you the right answer, and it can become a way of dismissing every external number while continuing to use internal ones that have no sample definition at all.

The specific limit with sample size is that a small sub-sample is not automatically wrong. Eighteen United States responses may be describing the market accurately, and a larger sample drawn badly can be worse. What the small number gives you is a reason to treat a single wave as weak evidence and to look at the direction across several waves instead.

The other limit is that show size is the variable almost nobody publishes. Response counts by country are common, response counts by size of show are rare, so you can usually establish whether your geography is represented and almost never whether shows of your scale are. That gap is real and no amount of careful reading closes it.

This week, take the last external benchmark figure your team quoted and find its sample paragraph. Write the response count, the geography and the recruitment method into the same document as the figure, and check the published research pages for the appendix that gives the count for your own market. If it comes to fewer than about 30 responses, treat the market figure as directional and move on.