How to Estimate Event Attendance From a Sample of Footage

A three-day festival is 30-plus hours of gate activity across six entrances. Filming, storing, and processing every single minute is often neither practical nor necessary. The honest alternative is to estimate event attendance from a sample: count representative windows of footage, then project the rest — carefully, with the assumptions written down and a confidence range attached. Done right, sampling gives you a full-event figure you can defend. Done carelessly, it is how "we think about 40,000" gets invented. This guide is about the difference, and how GateProof keeps the projection honest.

Key takeaways

• Sampling counts representative windows and scales up by how much of the event you actually observed.
• Every estimate carries a confidence interval (counting noise) and a confidence tier (how representative the sample was) — two different things.
• More coverage earns higher confidence; a short spot clip is marked indicative and never inflated into a confident day total.
• Monthly and annual projections are withheld when the footage cannot support them, because inventing them would be fabricating data.

Why sampling exists — and why it must be honest

You already accept sampling everywhere. A political poll asks 1,000 people, not the whole country, and reports a margin of error. Sampling event attendance is the same idea: measure part of the whole precisely, then reason about the rest with stated uncertainty. The danger is not sampling itself — it is sampling and then throwing the uncertainty away, quoting the projected total as if you had counted every head. That is the move that gets attendance numbers laughed out of a renewal meeting. GateProof samples openly and keeps the uncertainty visible in the Attendance Report.

What "representative" actually means

A sample only projects well if it looks like the whole. Fifteen minutes at the dead 3pm lull will under-count a festival; fifteen minutes during the headline rush will over-count it. Good sampling spreads windows across the shape of the event — the quiet open, the build, the peak, the drain — so the projection reflects a real day, not a cherry-picked slice. When you cannot film continuously, film short windows across the day rather than one long block at one time.

How to estimate event attendance from a sample

When your observed footage does not cover the full event, the engine does not simply guess the missing hours. It counts what it can, then scales by observed coverage — the fraction of the event actually on camera — and attaches a confidence interval around the result. That interval widens as the counts get noisier, so a dense, hard-to-count gate produces a wider band than a clean, quiet one. The maths behind the interval is deliberately conservative; if you want it in full, see what "accurate" really means: MAPE and confidence intervals, which lives on our methodology hub.

Two kinds of uncertainty, kept separate

Here is the honesty distinction most attendance figures blur. There are two different uncertainties in any projected total:

1. Counting noise — the natural variation in the count itself. This is what the confidence interval on the number captures.
2. Representativeness — whether your sampled windows actually stood in for the whole event. This is not something a confidence interval can fix, so it is reported separately as a confidence tier.

That is why a report can show a tight interval and a "low confidence — indicative" label at the same time: the arithmetic on the sample was precise, but the sample was too short to speak for the day. Treating these as one number is exactly how organizers get misled. Keeping them apart is what makes the estimate trustworthy.

A worked example, honestly

Say your Saturday market runs 09:00–17:00 across two gates, and you only managed to film four 20-minute windows at each — spread across open, mid-morning, lunch peak, and the wind-down. That is roughly 1.3 hours of an 8-hour day, or about 17% coverage. Sampling scales the observed crossings up by that coverage fraction and attaches a confidence interval to the result. Because you sampled across the day's shape rather than all at 3pm, the projection is defensible and lands in the medium tier — reported as a range, not a single figure. Film the full day at both gates instead and the same event earns high confidence with a tighter band. The report always shows which tier your footage bought, so nobody mistakes a sampled figure for a full census.

The confidence tiers, in plain terms

• High — a full event day counted end to end, or multiple observed days. Project with confidence.
• Medium — a partial but substantial window on the day. Usable, stated as an estimate with a range.
• Low / indicative — a short spot reading. Reported as a rough indication only; no confident day total, and no monthly projection at all.

The tier is not a marketing dial. It is the report telling you, honestly, how far your footage can carry the claim.

What sampling will not do

It will not manufacture precision you did not buy. A ten-minute clip cannot become a certified 40,000. A single day cannot become a defensible monthly footfall figure — that projection is simply withheld when the footage cannot support it, because printing it would be inventing data the count never contained. If you need a high-confidence full-event total, the answer is more coverage, not more optimism — which is exactly why a Festival package covers up to three days and six counting points.

Sample smart, then verify

The practical recipe: cover your busiest gates fully, sample the quieter ones across the day, and let the report show its tiers and intervals rather than hiding them. If you are capturing footage yourself, pair this with how to count event attendance with a phone. For the full landscape of methods, start at the pillar, how to count event attendance.

Want to see how your footage scores on confidence before you commit? Count your gates with a $199 Event Report at your next event and read the tier it earns.