Event promotion: online discovery and word of mouth overlap

The study covers U.S. adults who reported attending at least one eligible arts event during the 12 months before their July 2022 interview. Eligible events included live music, theater or dance; book readings, poetry or storytelling; and art exhibits. Elementary, middle and high school performances and online events were excluded. Movies were outside these discovery questions.^[1]

We analyzed 2,188 records with a confirmed attendance answer and a positive survey weight. Respondents were asked about ways they first heard about the events they attended over that year. A person could name several sources because they had attended several events, or because the sources overlapped in their account. The survey does not connect each source to a named event.^[3]

Here, “online” includes social media, websites and other online sources. “Peers” means friends, neighbors or co-workers. The questionnaire asked about family separately. A peer might also share a link online, so these categories describe answers to the questions, not mutually exclusive communication technologies.

The reference period falls roughly between July 2021 and July 2022, during the recovery from pandemic restrictions. These are historical U.S. arts-attendance results. Their practical value is in showing what a channel report can contain; they are not a current benchmark for a particular venue, country or event format.

An estimated 30.4% of confirmed attendees reported both online and peer discovery. Another 23.2% reported online discovery and explicitly answered No to peer discovery. Peer discovery with an explicit No to online accounted for 15.5%. A further 27.9% answered No to both, and 2.9% could not be fully classified because at least one answer was missing.

Each row below uses the same group of confirmed attendees. The percentages use survey weights; the respondent counts are the unweighted records behind them. The approximate 95% sampling intervals describe uncertainty from the sample design, separately from the missing answers.

The shared group is visible only when the two answers are read together. A report that lists online mentions and peer mentions in separate totals does not show whether they refer to separate people. In this study, they often did not.

“Neither” also has a specific meaning. It means No to these two source questions. It does not mean that the person attended without hearing about the event, or that event promotion failed. The questionnaire included other sources, and this analysis does not assign the neither group to any of them.

For a venue or creator, the useful reporting choice is to retain the combinations. You can then discuss the people who name both sources without hiding those who name only one. The table supplies no basis for dividing a campaign budget according to these percentages.

Yes. Among attendees with an observed Yes to online discovery, 56.5% also answered Yes to peer discovery. The approximate 95% sampling interval is 52.7% to 60.3%. Another 43.2% answered No to peers, while 0.3% had an unanswered peer item.

This comparison uses a smaller denominator: the 1,165 records with online discovery confirmed. It retains the three records whose peer answer was missing. People with an unknown online answer are outside this particular group, even if their peer answer is known.

The result changes the question an organizer might ask after seeing strong online discovery. How much of that audience also reports a personal connection? In this historical sample, more than half of the weighted online-Yes group also named friends, neighbors or co-workers.

That shared answer does not establish a sequence. It could describe an online post and a conversation about one show, or online discovery of one event and peer discovery of another. It also cannot distinguish an unpaid post from an advertisement. The next useful evidence would come from questions tied to your own named event.

The 44 incomplete records stay in the main denominator. We do not treat a refusal, a “don’t know” response or another missing-answer code as No. Instead, we ask how far each category could move if the missing discovery answers took either possible value while the known answers stayed fixed.

For the both-sources category, that gives a range from 30.4% to 33.1%. The lower end counts confirmed Yes/Yes records. The upper end also includes incomplete records whose existing answers still allow Yes/Yes. The same calculation is shown for every fully classified category.

These are logical ranges from missing answers, not confidence intervals. Each upper end is considered separately; the four upper ends are not a possible distribution that should add to 100%. The ranges address missing discovery items among confirmed attendees, not people who did not respond to the survey, uncertain attendance, or errors in recall.

The online-Yes group has fewer unresolved peer answers. Its overlap range is 56.5% to 56.8%, holding all known answers fixed. This narrow range does not remove sampling uncertainty: the separate 52.7% to 60.3% sampling interval still matters.

In a business report, this distinction helps keep two questions separate. One asks how much incomplete answers could change the classification of the observed sample. The other asks how uncertain the estimate is because it comes from a sample. A small missing-answer range is not a promise of precise prediction.

To apply the lesson, make the event and the observation period explicit. A short survey after a show, reading or exhibition can ask about that specific occasion. The following wording is an illustrative template, not the questionnaire used to produce the historical percentages above.

Start with: “Thinking about [event name] on [date], did you hear about it through any social media, website or other online source before attending?” Offer Yes, No, Don’t remember and Prefer not to answer.

Then ask: “Before attending this event, did you hear about it from a friend, neighbor or co-worker?” Use the same response options. Keep these as separate questions so a person can answer Yes to both, and retain a skipped question as unanswered.

This template asks whether someone heard about the event through a source, including possible reminders. If first discovery is the decision you need to understand, ask a separate question about the first remembered encounter and allow uncertainty. Do not quietly describe the broader heard-about answers as first-touch attribution.

Keep the survey response attached to one respondent and one event without collecting more identifying information than the task needs. If the same person can submit more than once, define how duplicates are handled before reporting the results. State who was invited, how many answered and when the survey was available.

A post-event survey describes the people who respond. It can miss non-attendees and people who ignore the invitation. It therefore helps explain reported discovery among respondents; it does not estimate how many extra tickets an advertisement sold. Keep the event’s ticket and campaign records under their existing measurement and consent rules.

Imagine a small venue collecting the two event-specific answers above. The following three response patterns are fictional examples of classification, not additional findings from the Census data.

The first attendee answers Yes to online and Yes to peers. Put that response in the shared group. The two Yes answers remain two source mentions from one respondent. A message from a friend containing a website link could fit both descriptions; the answers alone do not tell you whether that happened.

The second attendee answers Yes to online and No to peers. This is an online-only answer pattern for the two questions. The word “only” is relative to the pair: it does not rule out family, a poster or another source that the survey did not ask about.

The third attendee answers Yes to online and leaves the peer question blank. Keep that response in the online-Yes denominator, but leave the peer classification unresolved. Calling it online-only would turn an absent answer into evidence the person did not provide.

The same principle applies to No/No answers and records with an unknown online answer. Preserve what is known, identify what remains unanswered and use a denominator that matches the question. This gives the organizer a readable account of the responses without inventing a path from promotion to purchase.

Use the result to improve the next conversation about promotion. When a team compares online discovery with word of mouth, ask whether it is comparing source mentions or groups of people. An overlapping audience can be hidden inside two apparently separate channel totals.

For the next event report, show the shared group, the two source-only groups, neither and incomplete answers together. Keep the event name and dates beside the table. If you also report a percentage among online-Yes respondents, name that smaller denominator rather than letting it look like a percentage of all attendees.

An overlap finding can motivate questions about how event information is shared. For example, an organizer may want to learn whether people forward a useful event page to friends. That is a hypothesis for an event-specific follow-up, not a behavior established by this survey. The same distinction applies to a creator promoting a reading, performance or exhibition.

Decisions about campaign spending need additional evidence, including the actual activity, costs and outcomes being compared. The contribution here is narrower and immediate: a better account of who reported which sources before a team assigns a commercial explanation to the totals.

We used the native public files and 160 person-level replicate weights for the July 2022 Survey of Public Participation in the Arts, administered as a Current Population Survey supplement. Interviews took place July 16–25. The supplement selected one eligible adult in about half of sampled households and permitted proxy responses when the selected adult was unavailable. Discovery questions were in the randomly assigned Core 2 module.^[1][3]

The analysis keeps records with positive supplement weights, age 18 or older, and at least one Yes to the three eligible attendance questions. It does not infer attendance from a discovery answer or require every discovery question to be answered. The full supplement contains 13,251 positive-weight records; our confirmed-attendance group is 2,188. These are different populations within the file.

For a Core 2 population total, the Census instructions apply a factor of two to the supplement weight. We apply the same factor to the full and replicate weights. It changes population totals but cancels within the ratios reported here. Every replicate uses its own denominator.

Sampling variance follows the Census replicate-weight method: 4/160 times the sum of squared differences between each replicate estimate and the full-sample estimate. The reported intervals use a normal approximation and are pointwise, not a simultaneous guarantee for the whole table. Displayed intervals are intersected with 0%–100%; the download retains the original unbounded endpoints.^[4]

The analysis uses a conservative publication floor of 75,000 in the weighted domain, following the Census caution about small percentage-distribution bases. Both the confirmed-attendance and online-Yes domains exceed that floor. This is our adopted rule, not a claim that the Census mandates this exact suppression policy.

The NEA’s 2025 report presents separate discovery-channel percentages. Our contribution is to read two answers together, preserve the incomplete cases and separate missing-answer ranges from sampling uncertainty. The existence of overlapping channels was already implicit in the published totals; we make no first-ever claim.^[2]

The downloads contain the displayed aggregate results and a reproduction package with source references, calculation code and the full numerical output. They contain no individual respondent records, identifiers or survey weights. Influence IQ is responsible for this reanalysis and its interpretation.