Reading a disaster: method, and the timeline

Introduction

The first account of a crowd disaster is almost always wrong in some way that matters. At the Love Parade in 2010, early reports said people had fallen from a staircase; they had not, and the cause of death was crushed rib cages. After the Hajj in 2015, the Saudi authorities gave a toll of 769 and never updated it; the Associated Press counted 2,411. At Hillsborough it took until the inquests of 2016 for a jury to record that the behaviour of supporters had not caused or contributed to the deaths.

So reading a disaster well is a skill, and it is the skill this module teaches. You now have the tools: density and flow, the fundamental diagram, the mechanics of compression, the psychology of crowds, and Still’s DIM-ICE model. This lesson gives you a method for applying them, and a timeline of every incident the course could verify, so that you can see the patterns before you study the cases one by one.

The case template

Every case in this module is written to the same seven-part structure. Use it on any incident you read about, and notice which parts the available sources leave empty: the gaps are often the most informative thing about a report.

Two habits make the template work. First, keep the crowd’s behaviour and the conditions it was placed in apart, and read the second before judging the first. The panic myth (m3-l1) is so persistent that a report which describes “panic” without describing densities and flows has usually skipped the hard part. Second, write every number with its date and its source. Tolls change, capacities are recalculated, and timelines are revised.

Naming the mechanism

The course classifies incidents by the physical mechanism that killed people, using five categories from its fact sheet plus one of its own. A standing crush is compression in an upright crowd pressed against a barrier or a bottleneck. A progressive collapse, or pile, is what happens when people fall and others fall on top of them. A stairway fall-pile is a collapse on a stair, where the slope keeps adding people to the pile from above. Then there are fires that drive a crowd into its exits, and structural failures of barriers, walls or walkways. The sixth category, tear gas and locked exits, appears twice in the timeline and once in the case studies.

Helbing and Mukerji give the distinction that matters most for prevention: “In a classical crush, people are moving towards a physical bottleneck and die in front of its narrowest point. In a ‘crowd quake’, there is typically no systematic flow directions, but people are pushed around by fluctuating forces.” A crush is prevented upstream, by limiting what arrives at the bottleneck. A crowd quake is prevented by never letting the density get high enough for turbulence to start.

Mechanisms mix. At Hillsborough, the Taylor report describes victims held upright by pressure all round, and others who fell when a barrier broke and were pressed down by those behind. The category records the dominant mechanism, not the only one.

The timeline

The timeline below holds every incident in the course’s fact sheet that is recorded as verified or reported by a reputable secondary source: 28 incidents from 1943 to 2025. Each circle’s area is proportional to the number of deaths. Where the toll is contested, the circle shows the lower or official figure, it is hatched, and a dashed ring shows the highest figure on record.

What the timeline shows, and what it hides

Three patterns are visible at once. Religious gatherings and football grounds account for most of the deaths. Stairs, tunnels and ramps recur: Bethnal Green’s shelter stair, Stairway 13 at Ibrox, the al-Ma’aisim tunnel, the Love Parade’s ramp and Itaewon’s sloping alley. And the UK’s stadium incidents stop appearing after 1989: Hillsborough’s final report led to all-seater stadiums in the top two divisions by August 1994.

The third pattern needs care. The claim that there has been no fatal crowd crush at a UK football stadium since Hillsborough could not be verified, so the course does not make it. Two recent events show why the record is no reason for comfort: at the Euro 2020 final at Wembley about 2,000 people entered without tickets, and a review found it was “genuinely lucky” that things were not worse; at the O2 Academy Brixton in December 2022, two people died in a crush at the doors of a music venue.

Two kinds of uncertainty run through the data. The first is the toll. Mina 2015’s official figure and the independent tallies differ by more than a factor of three; at Prayagraj in 2025 the official figure went from 30 to 37 while a BBC Hindi investigation found at least 82; Phnom Penh’s official toll was revised down from 456 to 347. The second is the mechanism. For many recent incidents no published analysis says how people died, which is why so many recent circles are hollow.

Design, information, management

Keith Still’s DIM-ICE model was built from the analysis of incidents, and it gives the template’s last question its vocabulary. Still sorts disasters into “Design problems (too many people, not enough space), Information issues (sudden changes in the environment, poor or no information, confusing signage etc) or management problems (good design, good information, badly managed).” In a 2020 paper he and colleagues wrote that “the design element was the fundamental causality in over half of the incidents”, and that “After major incidents when facts are analysed the crowd is rarely the cause.”

A useful way to classify a failure is to ask a counterfactual for each letter. If the space had been big enough for the number of people who came, would it still have happened? If the people arriving had known what was happening ahead of them, would they have kept coming? If someone with authority had acted on what they could see, would it have stopped? A “yes, it would still have happened” on two of the three points to the third. Most of the cases that follow fail on more than one.

Each case study ends with a DIM-ICE matrix filled in from the inquiry findings. Red marks a failure that the cited findings say contributed; amber, a factor that played a part; green, one that worked or was not at issue. The cells are editable, so you can record your own reading next to the course’s.

How to read an inquiry report

Inquiry reports are long, and the order in which they are published matters as much as what they say. A few habits help.

• Note which stage you are reading. Hillsborough had an interim report in August 1989 that named “the failure of police control” as the main reason, a final report in January 1990 with 76 recommendations, an Independent Panel report in 2012, and inquests that concluded in 2016. Each later stage revised what came before.
• Separate findings from recommendations. A finding says what happened; a recommendation says what should change. Recommendations tell you what the authors thought was fixable, which is not the same as what was most important.
• Check the medical evidence against the narrative. The Hillsborough Independent Panel cast “significant doubt” on the assumption that every victim lost consciousness within seconds and died within minutes. That assumption had shaped earlier conclusions about what the emergency response could have achieved.
• Read criminal outcomes for what they are. A trial that ends without a verdict, as the Love Parade trial did in 2020, is not a finding that nobody was at fault. A grand jury that declines to indict, as at Astroworld in 2023, answers a criminal question, not a safety one.
• Note what is still open. In Itaewon’s case a special commission is due to report by September 2027. Anything you write before then is provisional.