Survey response rate reporting and the figure most studies leave out

A benchmark slide arrives in the pre-show pack carrying one footnote: n = 466. Somebody in the room asks where the 466 came from. The honest answer is that 466 companies answered, and nobody present knows how many were asked.

Survey response rate reporting is the practice of publishing that second number, and it is the one most exhibition industry research leaves out. The sample size gets printed because it looks like evidence. The denominator gets left off because it usually looks small.

Why does the sample size on its own tell you so little?

Five hundred responses is five hundred responses whether it came from six hundred invitations or from two hundred thousand. In the first case you have heard from most of the population and the answers are close to a census. In the second you have heard from a quarter of one per cent, and the 199,500 who ignored the email are the interesting group, because there is no reason to assume they resemble the ones who replied.

Work the simple version. A study reports 500 responses from an invitation list of 20,000. That is a response rate of 2.5 per cent. The 19,500 who did not answer are 39 times the size of the group you are reading, and every conclusion in the report rests on the assumption that those 39 people per respondent would have said roughly the same thing.

That assumption fails in a direction you can usually predict. Organisers whose last edition went well answer questions about how the last edition went. People with a strong opinion about a pricing change answer surveys about pricing. The bias sits in the gap between respondents and invitees, and the response rate is the only published quantity that tells you how wide the gap could be. Which way that gap leans, and by how much, is the separate question of self selection rather than this one.

Six response rates, one survey file

The American Association for Public Opinion Research publishes the reference document here, Standard Definitions, whose tenth edition appeared in 2023. It gives operational formulas for response, cooperation, refusal and contact rates, and it is explicit that there is no single response rate:

In calculating and reporting outcome rates according to the rules and formulas below, researchers must precisely define the rates used. For example, a statement that "the response rate is X" is unacceptable. One must report exactly which rate was used, such as "Response Rate 2 was X."

AAPOR (2023) credits the lineage to the 1982 Special Report on the Definition of Response Rates from the Council of American Survey Research Organizations, and extends its formulas. Six response rates are defined, ranging from the one that yields the lowest figure to the one that yields the highest, and the difference between them is how partial responses are treated and what you assume about cases whose eligibility you never established.

Run a post-show exhibitor survey through them. You mail 20,000 addresses from your exhibitor contact file. You get 380 complete responses and 120 partials. You get 240 explicit refusals, 460 hard bounces and out-of-office replies from people who never engaged, and 200 other dispositions such as duplicate submissions you threw out. That leaves 18,600 addresses that did nothing at all, and because your file is three years old you have no idea how many of those are still live mailboxes belonging to eligible exhibitor contacts.

RR1, the minimum, is complete responses over everything: 380 divided by 20,000, which is 1.9 per cent.

RR6, the maximum, counts partials as responses and assumes none of the unknowns were eligible: 500 divided by 1,400, which is 35.7 per cent.

Same file. Same fieldwork. One point nine per cent or thirty five point seven per cent, and both are correctly computed. RR3 sits in between by estimating what proportion of the unknown cases were eligible. Assume half of your 18,600 stale addresses were live and eligible, and you get 380 divided by 10,700, or 3.6 per cent. AAPOR (2023) warns against picking that proportion to flatter the result and requires the basis for the estimate to be stated in detail.

This is why "the response rate was 36 per cent" is an unreadable sentence. The number is defensible and it is also nineteen times the minimum, and the reader has no way to tell which of those two facts they are looking at.

What the exhibition industry actually publishes

The UFI Global Exhibition Barometer, 37th edition, published in July 2026 from a survey concluded that June, is the closest thing this industry has to a standing measurement instrument. It publishes a full appendix listing replies per country, 466 in total across 59 countries and regions, broken down to single-digit precision: 178 from Europe, 113 from Asia-Pacific, 62 from North America, 57 from Central and South America and 56 from the Middle East and Africa. Those five figures sum to 466, which is more transparency about composition than most published research offers.

There is no response rate anywhere in its 142 pages, and no statement of how many companies were invited. UFI is candid about the consequence in its own remarks, writing 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". That is a fair caveat, honestly placed. It is also not a substitute for the denominator, because a reader who wanted to size the gap still cannot.

Explori, which describes its work as running through "partnerships with UFI, IMEX and other organisations setting the standard for global event measurement" (Explori, 2026), sits in the same position: benchmark databases built from event feedback surveys where the per-event response rate is the property that decides whether the benchmark is a population estimate or a portrait of the enthusiastic.

None of this makes the research bad. It makes the research incomplete in a specific, fixable way, and the fix costs the publisher one line.

How do you ask a publisher for the denominator?

Directly, by email, in one paragraph, before you quote the number to a board. Ask for four things.

• The invitation count. How many distinct cases were contacted, and were they a list, a panel, or an open link that anyone could forward.
• Which rate. Name AAPOR's scheme and ask which of RR1 through RR6 was computed, or ask for the completes, partials, refusals and unknowns so you can compute it yourself.
• The disposition table. AAPOR (2023) asks that a table of final disposition codes be prepared and made available on request. Most publishers have it in the field report even when it is absent from the public deck.
• Whether an incentive was offered, and to whom, since an incentive changes who answers and not only how many.

If the answer to the first question is that there was no list, because the survey was promoted through a newsletter and a conference session, then no response rate exists and the study is a self-selected sample. That is a legitimate design for some questions and it should be labelled, which is where the sample definition itself has to be read before the headline is quoted at all. The mechanics of how the invitation reached anyone, whether through an association list, a panel or an open link, belong to the recruitment question, and they decide whether a denominator was ever knowable.

Reporting your own post-show survey

The reason to care about this beyond citation hygiene is that your own exhibitor and visitor surveys have the same defect, and yours goes into documents that renew contracts.

Publish three numbers together on every survey page in the post-show report: the invitations sent, the completed responses, and the rate with its definition named. If you mailed 4,100 exhibitor contacts across 612 exhibiting companies and got 214 completed responses from 189 companies, say so. The company-level rate of 189 over 612 is 30.9 per cent and the contact-level rate of 214 over 4,100 is 5.2 per cent. Both are true, they differ by a factor of six, and which one belongs in the report depends on whether the unit of analysis is a company or a person.

Then hold the definition steady between editions. A survey that moved from an open link on the exit page to a targeted mailing will show a satisfaction shift that has nothing to do with satisfaction, and a trend line built across that change is measuring the instrument. Freezing the method matters more than optimising it, at least until you have three editions on the same basis.

Where this stops

A response rate bounds the risk of non-response bias. It does not measure it, and treating a high rate as proof of accuracy is its own error.

A survey with a 45 per cent response rate whose respondents are all large organisers can be further from the truth about a portfolio of small shows than a 6 per cent survey that happened to reach a balanced spread. The rate tells you how much room there is for the respondents to differ from the population. Whether they actually differ is a separate question, answered by comparing respondents against known population characteristics such as show size, region and sector, which requires the publisher to hold those characteristics on the frame in the first place.

That comparison is the thing worth asking for after the denominator. If a study can show that its 466 respondents match the regional and size distribution of its invitation list, a low rate stops being alarming. If it cannot, a high rate is not much comfort either.

This week, take the last industry statistic you put in front of a commercial decision and find its denominator. If the report does not carry one, email the research contact and ask for the invitation count and the disposition table, then write the answer, or the absence of one, next to the number in your own file. Keeping that annotation with the figure is the cheapest piece of measurement standards discipline available to a team of one.