Lessons from the World Trade Center evacuation

Introduction

The evacuation of the World Trade Center towers on 11 September 2001 is unusually well documented. That is not because it was typical, but because hundreds of survivors could later describe, in detail, what they did and why. Their accounts overturn most of what popular culture assumes about people in emergencies.

There was no stampede for the exits. People talked, telephoned, gathered their things, helped colleagues, and then walked down dozens of flights of stairs, often standing still for long stretches. This lesson reads the evidence as an evacuation engineer would, with the two clocks of lesson 1 in mind. It is about how people behaved, not about the attack; the victims are not discussed here.

Two studies

The United States National Institute of Standards and Technology (NIST) examined the occupants’ behaviour and egress in its report NCSTAR 1-7, by Averill and colleagues.

In the UK, Ed Galea’s fire safety engineering group at the University of Greenwich led the HEED project, with the universities of Ulster and Liverpool. Funded by the EPSRC with £1.6 million, it ran from 1 September 2004 to 30 April 2008. The overview paper by Galea and colleagues in 2012 reports interviews with 271 survivors and almost 6,000 pages of transcripts. The project’s own introduction page says “almost 300” interviews; the two figures contested probably count different things, and this course uses the paper’s 271.

Before moving

The first finding is the one lesson 1 predicted. In WTC 1, NIST found a median time to start evacuating of 3 minutes for occupants on floors up to 76, and 5 minutes on floors 77–91, near the impact region. Among its sample of 440 survivors, 70% talked to others and 46% gathered personal items before leaving; NIST’s tables also suggest that about 30% helped others and about 23% searched for others, although how those categories map onto the questions is less certain.

HEED’s interviews let Galea’s team put numbers on the delay for 119 people. 13% responded within a minute, 78% took between 1 and 8 minutes, and 9% took more than 8. On average, people carried out four tasks before setting off: about 1.5 seeking information and 2.5 actions. “Collect Items” was reported 143 times, 49% of all action tasks.

These are not signs of panic. They are the actions of people trying to understand an ambiguous situation and to leave with what they would need, in a building they had every reason to think was safe. They are exactly the “recognition” and “response” phases of pre-movement, and they line up with the psychology of module 3’s panic-myth lesson: the danger is delay, not flight.

On the stairs

Once people were moving, the stairs set the pace. HEED’s adjusted mean descent speed on the stairs was 0.29 m/s, slower than a quarter of the free walking speed of 1.34 m/s you met in module 1. At that speed, 100 m measured along the stair takes nearly six minutes (our arithmetic).

The interviews also recorded when and why people stopped. Of 388 reported stops on the way down, 43% were caused by congestion. The stair was full; people waited for the people in front, and at every floor, for the people merging in. That is lesson 2’s landing problem, repeated over dozens of storeys.

What it means for design and messaging

Put the findings together and they make a short list of implications for anyone designing or managing a tall building, or any building where people must travel far to safety:

1. Expect delay. Model pre-movement as a distribution with a long tail, not a constant. Most people take minutes; some take far longer.
2. Information shortens delay. People spent their first minutes finding out what was happening. Clear, specific, early instructions attack the largest block on the RSET clock. Module 3’s collective resilience lesson explains why honest information works.
3. Design the stairs for capacity and merging. Stair width and the floor–stair connection matter more than travel distance or walking speed, because the stairs are where people stand still.
4. Calibrate models to real behaviour. A simulation that starts everyone at once and lets them walk at free speed will be badly optimistic. Data like HEED’s exist so that models can do better.