How Accurate Are Rally Estimates?

Crowd estimates at rallies, protests, marches, and vigils are often disputed because different observers use different methods, assumptions, and time windows. One source says 10,000, another says 50,000, and neither figure is very useful unless someone can explain how it was reached.

The real question is not which number is right—it is whether the estimate can be trusted at all.

Recent and historic examples show how large these gaps can become:

• On 3 August 2025, the Sydney Harbour Bridge "March for Humanity" produced widely different figures. NSW Police initially estimated about 90,000 attendees, ABC reported more than 100,000, organisers said the total was closer to 300,000, and The Guardian cited an independent expert estimate of roughly 225,000 to 300,000 based on drone footage and density analysis.
• On 16 June 2019, a major Hong Kong anti-extradition protest produced another large gap: organisers said nearly 2 million people attended, while police said 338,000 were counted on the agreed route at the march's peak.
• On 15 February 2003, the London anti-Iraq war march produced one of the best-known organiser-authority splits: police said at least 750,000 people attended, while organisers put the figure at 2 million.

These examples do not prove that one side is always right and the other always wrong. They show that the reliability of a crowd estimate depends on method, timing, and transparency.

Crowd estimation is never perfectly accurate. Several factors introduce systematic errors:

• Observer position. Someone at one point in a march sees different density than someone at another point.
• Time variation. Peak crowd is rarely constant. A crowd of 20,000 at noon may have grown to 30,000 by 2pm.
• Density assumptions. Calling a crowd "packed" is subjective. Different observers may assume very different people-per-square-meter densities.
• Invisible or split attendance. Some participants leave early, others arrive late. Some may be in side streets or less obvious gathering areas.
• Movement and routes. A marching crowd is not static. People join and leave continuously.
• Reporting incentives. Organisers may report higher numbers, authorities lower ones. Both have reasons to shape the estimate in their direction.
• Visibility into check-ins. Many people participate but do not check in or register. The visible signal always undercounts actual presence.

Each of these factors can shift an estimate by 10%, 30%, or more. None of them makes an estimate useless, but all of them matter for understanding what an estimate really means.

Accuracy in crowd estimation does not mean false precision.

An estimate of "approximately 25,000 with a margin of error around ±20%" is more accurate than an estimate of "50,000" with no acknowledgment of uncertainty.

Useful accuracy includes:

• Transparency about method. How was the estimate reached?
• Stated assumptions. What density was assumed? What time of day?
• Margin of error. Is this a point estimate or a range?
• Definition of what was counted. Peak presence or total attendance?
• Awareness of what was not counted. Remote supporters, satellite events, or people who arrived too late to be observed?

A crowd estimate is accurate when someone else can understand how it was reached and evaluate whether the method was sound.

Some estimation approaches are less reliable than others:

• Single-photo estimates. A photograph captures one moment and one angle, not the whole event.
• Visual guessing with no structure. "That looks like 30,000 people" is less reliable than a zone-by-zone area and density count.
• No time annotation. If an estimate does not specify when during the event it was measured, it is incomplete.
• Mixing peak and total attendance. Confusing these two metrics introduces immediate error.
• Single-method estimates for complex events. A march that moves through the city cannot be accurately estimated with one number measured at one location.
• No secondary validation. Estimates tested against multiple methods (aerial counts, entrance/exit tallies, participant check-ins) are more reliable than estimates from one source alone.

When you hear a crowd figure reported, ask:

1. Can the method be explained? If not, treat the number skeptically.
2. What time of day was it measured? Different times produce different peak figures.
3. Is it peak presence or total attendance? These are often confused and can differ by 50% or more.
4. What area was measured? For a marching crowd, knowing the length and route is essential.
5. What density was assumed? Dense, moderate, or sparse changes the result dramatically.
6. Are there independent confirmations? Organisers, police, media, and third-party observers often produce different numbers. If they align roughly, that increases confidence.
7. What was not counted? Were satellite locations, remote participants, or late arrivals included?

A trustworthy estimate answers most of these questions. One that answers none of them is just a claim.

For many open public events, a range is more defensible than a single number.

If a route-based estimate suggests 18,000 to 24,000 participants depending on density assumptions, publishing that range is usually more honest than claiming exactly 21,347. A range tells readers that the estimator understands uncertainty rather than hiding it.

This is especially important for:

• moving marches
• multi-location actions
• long events with changing turnout
• crowds measured from incomplete vantage points

Different crowds require different estimation approaches. None is perfectly accurate, but some are stronger than others for particular situations:

Area and density:

• Useful for static crowds (rallies, vigils)
• Requires clear space boundaries and density assumptions
• Moderately accurate if density is consistent; less accurate if crowd is mixed

Entrance/exit counting:

• Useful for ticketed events or controlled entry points
• Very accurate if all entrances and exits can be monitored
• Less useful for open, unticketed public events with multiple entry points

Aerial or satellite imagery:

• Useful for large outdoor events
• More objective than visual estimation
• Can count pixels or use density algorithms
• Requires image quality and clear event boundaries

Section-by-section:

• Useful for marches and routed events
• Requires observers at multiple points along the route
• More accurate than single-point observation
• Labour-intensive

Direct check-ins and participation signals:

• Useful for events where participants can register or check in
• Produces a hard count rather than an estimate
• Cannot be more accurate than the participation signal itself (some people never check in)
• Works best when combined with other methods

Composite methods:

• Combines several of the above approaches
• Produces more reliable results than any single method
• More labour-intensive but produces stronger evidence

One major source of inaccuracy in crowd estimates is that the observer and the counter are usually the same person or organisation. Organisers count one way. Police count another. Media outlets repeat whichever estimate they hear first.

An independent attendance signal—where participants themselves contribute information about their presence—changes that.

When participants can check in anonymously at an event, they are not guessing. They are reporting their own presence. That direct signal cannot replace visual estimates or area counts, but it can validate them.

If visual estimates suggest 20,000 people and only 5,000 check in, that is useful information. It might mean the visual estimate is too high, or it might mean participation signals were not widely publicized. Either way, comparing the two methods strengthens the evidence base.

"Exactly 23,847 people attended" sounds more precise than "approximately 25,000, probably between 20,000 and 30,000."

But precision is not the same as accuracy. A wrong number stated with decimal precision is still wrong. An approximate number with a clear method and a margin of error is far more trustworthy.

Good crowd estimates sacrifice precision for accuracy. They acknowledge uncertainty rather than hiding it behind an impressive-sounding figure.

Different types of events pose different accuracy challenges:

Static rallies:

• Area and density methods work well
• Density is often fairly consistent
• Peak attendance is clearest
• Moderately accurate if space is clearly defined

Marching protests:

• Section-by-section methods work best
• Requires multiple observation points
• Crowd changes as it moves
• Accuracy depends on how well route is documented

Festivals and multi-stage events:

• Zone-based counting needed
• Attendance can vary dramatically between stages and times
• Floating populations make it hard
• Composite methods usually required

Viral or decentralized events:

• Unmapped locations are often missed
• Hard to distinguish "participated" from "happened to be nearby"
• Low accuracy unless participants actively check in
• Requires transparency about coverage gaps

A crowd estimate that says "We counted approximately 15,000 people with an estimated margin of error of ±3,000" is more credible than one that says "15,000 attended" with no caveats.

Uncertainty is not weakness. It is honesty about the limits of observation.

When you acknowledge what you do not know, people trust what you do know more.

For organisers, journalists, and researchers who want more accurate crowd estimates:

1. Use multiple methods. Area counts, entrance tallies, check-ins, and aerial observation all reveal different truths.
2. Map the event. Document location, route, timing, and structure clearly.
3. Time-stamp observations. Note what time each estimate or count applies to.
4. Separate peak from total. Be explicit about whether you are measuring maximum presence or cumulative participation.
5. State assumptions. Say what density was assumed, what area was measured, what was included and excluded.
6. Invite independent validation. Support check-ins, encourage participant reporting, make methods transparent.
7. Revise estimates as evidence arrives. Better data may change the number. Updating transparently strengthens credibility.

• ABC News on the Sydney Harbour Bridge march, 3 August 2025: https://www.abc.net.au/news/2025-08-04/pro-palestine-large-crowd-perilous-on-bridge-march-police-say/105607802
• The Guardian on the Sydney Harbour Bridge march, 3 August 2025: https://www.theguardian.com/australia-news/2025/aug/05/we-know-the-sydney-harbour-bridge-march-against-the-killing-in-gaza-was-huge-but-just-how-big-was-it
• CNBC on the Hong Kong protest, 16 June 2019: https://www.cnbc.com/2019/06/16/hong-kong-protesters-demand-top-official-quit.html
• The Washington Post on the Hong Kong protest, 16 June 2019: https://www.washingtonpost.com/world/large-scale-protests-return-to-hong-kong-despite-suspension-of-extradition-bill/2019/06/16/7ea7f9c6-8ee0-11e9-b6f4-033356502dce_story.html
• CBS News on the London anti-war march, 15 February 2003: https://www.cbsnews.com/news/massive-anti-war-outpouring/
• London Museum on the 2003 anti-war march: https://www.londonmuseum.org.uk/collections/london-stories/stop-war-londons-largest-ever-protest/

Rally Tally cannot replace careful observation and density estimates. What it does is add a new data source: direct participant check-ins.

When a participant checks in to an event through Rally Tally:

• It is a hard count, not an estimate
• It is anonymous, so it does not require personal identification
• It is geographically registered, so it ties to a specific event location

Combined with visual observation, area estimates, and other methods, check-in data makes the final crowd estimate stronger because it includes what participants themselves reported about their presence.

The next time you see a crowd figure reported with high confidence and no method attached, be skeptical.

The next time you want to report on a crowd more accurately, prioritise transparency over impressive-sounding precision.

For public events that matter—protests, rallies, marches, gatherings—better accuracy comes from:

• Clear methods explained in plain language
• Multiple independent estimates compared
• Honest acknowledgment of margins of error
• Documented event structure and timing
• Direct participation signals when available
• Transparency about what was counted and what was not

If you want to improve accuracy at your next event, explore Rally Tally's tools for mapped event tracking and anonymous attendance check-ins.