Safety Science 38 (2001) 157±182 www.elsevier.com/locate/ssci Quanti®cation of behaviour for engineering design standards and escape time calculations D.A. Purser *, M. Bensilum Fire Safety Engineering Centre, Building Research Establishment, Garston, Watford WD25 9XX, UK Abstract Occupant behaviour in ®res depends upon interactions between the occupants, the building and the developing ®re. Although reasonable calculation models exist for the estimation of movement time (the time required for occupants to ¯ow out of the building), time required for behaviours taking place before the movement phase, collectively known as pre-movement time, are poorly described and quanti®ed. A series of monitored evacuation studies and investigations of ®re incidents in a range of dierent building types is described. Strategies for the application of behavioural data to design standards and escape time calculation methods are discussed. # 2001 Elsevier Science Ltd. All rights reserved. Keywords: Occupant; Behaviour; Human; Evacuation; Escape; Fire engineering 1. Introduction Fire safety depends upon the performance of a system consisting of the building and its services, the construction materials and contents that might become involved in a ®re, and the behaviour of the occupants. An inherently ®re safe system relies as little as possible on the positive actions of the occupants, while an unsafe system relies for most of its safety on the actions of the occupants. One function of the building and its systems is therefore to protect the occupants from ®re hazards. For scenarios involving the escape of occupants from a ®re, survival depends upon the outcome of two parallel processes: 1.1. The developing hazard from the ®re This incorporates ignition, ®re growth and the spread of ®re and ®re euent. These depend upon a range of variables, such as the nature and disposition of the * Corresponding author. Tel.: +44-1923-664936; fax: +44-1923-664910. E-mail address: purserd@bre.co.uk (D.A. Purser). 0925-7535/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved. PII: S0925-7535(00)00066-7 158 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 ®re load, potential ignition sources, the reaction to ®re properties of the lining materials and contents, the height and ventilation of the compartment and the nature of the ®re euent. The actions of occupants and the provision of passive containment and active smoke extraction or suppression systems also aect the rate of development and extent of the hazard from a ®re. Assessment of these processes for any particular scenario is aimed at calculating the time when an occupant would receive an incapacitating exposure to ®re euent. 1.2. The process by which occupants escape This depends upon detection, the provision of warnings, response to warnings (pre-movement time), occupant pro®le (such as age and physical and mental ability, sleeping or waking, population density) subsequent pre-egress behaviour (such as seeking information, collecting belongings, choosing an exit and other activities), egress (including way®nding, movement towards an exit, crowd ¯ow and other factors), design of escape routes, exit numbers and widths, and the psychological and physiological in¯uence of exposure to heat and smoke on escape behaviour. Assessment of these processes for any particular ®re scenario is aimed at calculating the time required for escape. Fire safety engineering design relies upon estimation of these critical time dependent features. 1.3. Major phases of evacuation For the estimation of time dependent features of evacuation, the behaviour of occupants is considered in terms of two main categories of processes. In the British and International Organisation for Standards ®re safety engineering (FSE) standards BSI DD240 (British Standards Institution, 1997) and ISO TR13387-8 (International Organisation for Standardisation, 1999). These are de®ned as pre-movement processes and movement (or travel) processes. These are de®ned as follows: 1.3.1. Pre-movement processes Pre-movement processes begin at an alarm or cue and end when travel to an exit begins. There are two components: Recognition (this begins at an alarm or cue and ends with the ®rst response): during this phase occupants continue with pre-alarm activities (e.g. shopping, sitting, eating, watching football) and Response (this begins at the ®rst response and ends when travel to an exit begins): during this phase occupants carry out a range of activities (e.g. stopping machinery, securing money or other risks, gathering children and other family members, investigating the situation, way®nding, alerting others, ®ghting ®re). 1.3.2. Movement processes Movement processes begin when travel to exits begins, and end when occupants reach a place of safety or leave the building (where evacuation is appropriate). D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 159 These de®nitions obviously represent a simpli®cation of the complex and varied sequences of actions that individual occupants can engage in during ®res. It is important to recognise that occupants can engage in a considerable range of activities involving movement during the ``pre-movement'' period. In this context, the term ``movement processes'' implies deliberate movement towards exits or through escape routes with the intention of leaving the building or reaching a place of safety. Behaviours occurring during the pre-movement phase may involve occupants moving around for other reasons, such as shopping or seeking information. In this context, ``recognition'' indicates that by the end of this period, an occupant or a group of occupants have recognised the need to cease normal activities and respond to the developing emergency. ``Response'' indicates the variety and sequence of activities engaged in to deal with (cope with) the developing emergency, but before the action of evacuating via a chosen escape route is begun. It is also important to recognise that these processes can be analysed both in terms of individual occupants and populations of occupants in various locations within a building. For each individual occupant, it is possible to measure (or de®ne in an evacuation calculation) a pre-movement and movement time. For a group of occupants within an enclosure, pre-movement and movement times may be expressed as average values, or in terms of time distributions (depending upon the complexity of the analysis). Where occupants are distributed in dierent enclosures within a building, it is possible to determine dierent pre-movement and movement time distributions for dierent enclosures. With regard to the analysis of movement processes, it is necessary to consider the time taken for occupants to travel to enclosure exits leading to escape routes and the patterns of occupant ¯ow through the escape routes. In multi-enclosure occupancies where the occupants are widely distributed (such as hotels or apartment blocks) pre-movement time is often the major determinant of evacuation time, especially the pre-movement times of the slowest occupants to respond. Travel distance also becomes relatively more important. In crowded situations, such as busy shops or theatres, evacuation time depends mainly upon the premovement times of the ®rst occupants to respond and the movement times of the whole occupant population. The ®rst occupants to respond form the fronts of the queues at the exits and determine the time from alarm to queue formation. After this, time to clear an occupied enclosure depends mainly upon exit route choice and the ¯ow capacities of exit routes. In prescriptive design, it is normal to discount the largest exit on the assumption that one exit may be blocked by ®re. The remaining exits should then provide sucient ¯ow capacity to enable any enclosure to be cleared within a certain period of time (for example 2.5 min in the UK), providing the evacuating population is distributed evenly to these exits. In practice, occupants tend to favour certain exits or escape routes over others, so that design clearance times may not be achieved (see Section 3.3.6). Reasonable calculation models exist for the estimation of movement time (Pauls, 1995; Nelson and MacLennan, 1995). Pre-movement times are poorly described and quanti®ed. This is especially true when the design of warning systems and emergency management strategies are inferior, so that occupant response on any particular 160 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 occasion is likely to be slow and unpredictable. For evacuation calculation times to become a viable component of ®re safety engineering, it is, therefore, vital that a database of pre-movement times and pre-movement time distributions is obtained for a variety of occupancy types and a variety of building design and ®re safety management strategies. 2. Brief description of methods used for behavioural studies Research at FRS (the Fire Research Division of BRE), includes analysis and quanti®cation of occupant behaviour so that safety implications and escape time can be predicted for a range of occupancies and designs. Evacuation studies have been conducted using video and questionnaire analyses for a range of occupancy types (Purser and Raggio, 1995; Purser and Bensilum, 1999). Fire investigations. involve studies of occupants' experiences and responses to developing scenario (Purser, 1994; Purser et al., 1998). The results are used for the development of calculation methods and computer models for use in ®re safety engineering design; also in the development of FSE standards (BSI DD240 and ISO DIS 13387-8). The following sections include brief outlines of a number of studies illustrating the behaviours occurring during evacuations and providing some examples of the quantitative pre-movement and movement time data and data distributions obtained. Of particular interest was the nature and timing of pre-movement and movement behaviour for occupants in different types of occupancies and dierent settings, and the in¯uence of cues, warnings and ®re safety management on evacuation behaviour and evacuation times. 3. Results of studies of occupant behaviour in ®re Traditional FSE calculation methods for means of escape are very simplistic and tend to assume that as soon as ignition occurs, all the occupants immediately disperse to the nearest exits. In reality, occupant behaviour is much more complex, and heading directly to the exits often comprises the smallest part of the time required for escape (Proulx and Sime, 1991; Sime, 1994; Sime, 1998). These behaviours are illustrated by reference to a few examples of occupant behaviour in both real ®re emergencies and evacuation studies conducted by researchers of the Fire Safety Engineering Centre, BRE in a range of building types. Occupant behaviour in evacuation studies is considered to be similar to that in real ®re emergencies in which the majority of the occupants evacuate in response to alarms without coming into direct contact with ®re and smoke (provided they are not aware in advance that an evacuation is planned). 3.1. Real ®re incidents 3.1.1. Clothing store Table 1 summarises the sequence of events during a ®re in a clothing store (Purser, 1998) captured by security camera recordings. The ®re occurred during the afternoon D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 161 Table 1 Sequence of events in clothing store ®re. Reproduced by permission of BRE Ltd Time from ignition (min.s) Occupant behaviour 0.19 Fire approximately 0.5 m ¯ame height. Customer sees ®re and warns shop assistant who investigates and goes to fetch ®re extinguisher Assistant ®ghting ®re with extinguisher, ¯ame height approximately 1 m, ®re quite large, assistant fails to extinguish and moves away All this time people are entering the shop, passing the ®re, shopping, and waiting at the checkout to pay for goods. Shop ®lling with smoke, people reluctant to leave shopping People evacuating through thick smoke Sta evacuating A few people occasionally re-enter near doorway Front doors shut from outside 1.19 0.19±3.30 3.30 4.00 4.15 4.00±5.00 6.00 and the video switched between four cameras, one inside the store near the ®re, one on the opposite side of the sales ¯oor to the ®re, one showing the checkouts, and the fourth showing the main front exit to the street. There were approximately 20±30 customers in the shop and ®ve sta members. Most occupants were in the front area at the checkouts, but a number were dispersed in the sales areas, occasionally in the vicinity of the ®re. During the period of the ®re, a number of people left or entered the store carrying out normal shopping activities. There was no sound track on the video so it is not possible to be certain that an alarm sounder was triggered, or when this might have occurred. It is likely that a store such as this would have had at least a manually triggered sounder alarm system, and that this would have been set o by sta discovering the ®re. Even if this was not the case, it is evident from the video that both sta and customers were aware of the ®re throughout most of the incident. In this example, the occupants were very slow to respond, but all left eventually without injury. 3.1.2. Department stores Two ®res in department stores that resulted in fatalities were the Manchester Woolworth's ®re (Home Oce Fire Department, 1980; Sime, 1994; Purser, 1998) and the Chester®eld Littlewoods ®re (Derbyshire Fire and Rescue Service, 1993; Purser, 1998; Purser, 1994; Fig. 1). The Woolworth's ®re involved mainly furniture while the Littlewoods ®re involved mainly clothing. In both cases rapid ®re growth occurred on an upper sales ¯oor containing restaurants and sales areas. Both resulted in large numbers of people being trapped in toxic smoke before they could evacuate. The smoke from the furnishings in the Woolworth's ®re was highly toxic and caused a number of deaths. In the Littlewoods ®re, the smoke was dense but of a relatively low toxicity so that most people escaped without serious injury, although there were two deaths, mainly resulting from smoke inhalation and carbon monoxide poisoning. There are a number of similarities between the behaviour of occupants in these two incidents. Both cases involved occupants in restaurant areas adjacent to sales areas, and in both cases, problems occurred with the evacuation of these occupants. 162 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Fig. 1. Chester®eld Littlewoods Department Store Ð Fire Floor. Reproduced by permission of BRE Ltd. In the Woolworth's ®re, occupants were unaware of the ®re during the early stages. The alarm was raised by a painter who alerted the management and shouted to occupants of the restaurant who had not seen the ®re. It was reported that some occupants of the restaurant were slow to react and some were reluctant to leave quickly when the alarm was ®rst raised. The 100 or so occupants of the restaurant eectively had only one exit available to them (the Oldham Street stairway), a situation that also occurred during the Littlewoods ®re. The rapid ®re growth resulted in rapid smoke logging and obscuration of the exits, so that some occupants became trapped in smoke. Based upon the nature of the materials involved and the results of full-scale tests conducted at BRE, this smoke would have been highly toxic, resulting in rapid intoxication. The behaviour of occupants during the Littlewood's store ®re has been investigated in some detail by the author (Purser, 1994; Purser, 1998). The store is on a sloping site with the front at ®rst ¯oor level and the rear at street level. The main sales area contained clothing on racks. On the morning at the time of the ®re (10:00 hours), there were around 35 sta in the building. On the ®re ¯oor, there were approximately 17 sta and 70 customers. There was one assistant and only a few shoppers on the sales ¯oor area, but the restaurant at the front of the building was quite full. The diners consisted largely of retired people and young mothers with children. A few examples of the responses of occupants illustrate some of the range of behaviours carried out and the time needed for their execution. The ®re was started deliberately in polyester coats on a wall rack. It was discovered almost immediately (¯ames approximately 1 foot high) by two female shoppers who shouted and attracted the attention of the sales assistant on the till. The women then left by the rear doors and walked round to the front of the store, arriving just before the ®re brigade (walking time estimated at approximately 2 min). The store was already D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 163 completely smoke-logged and people were climbing out of the front windows onto a ®rst ¯oor ledge. Inside the store, the sales assistant had gone to a manual call point, started the alarm and instigated the emergency procedure. She then went to a set of triple doors, the only available exit from the restaurant, and held them open to assist people through. The ¯oor supervisor, who came out of her oce and went to investigate, was accosted by a male customer demanding a ®re extinguisher. She went for an extinguisher and they attempted to ®ght the ®re while the customer's wife waited. The ®re grew very quickly along a wall of clothing on racks, and was by now too big to extinguish. At this point, a number of sta members descended to this area of the ®re ¯oor from the ¯oor above. They proceeded immediately to the exits. The customer's wife then suggested that they should leave the building. The male customer, his wife and the ¯oor supervisor then left the building by the rear (street level) exits. Customers in the sales area during the period of early rapid ®re growth remarked that diners at the restaurant were slow to move. Meanwhile, the diners in the restaurant had heard the alarm and were responding. An adult family group included a brother and sister and their elderly mother. The sister heard the alarm, saying that it was probably a ®re drill, but her brother saw the ¯ames and said they must leave. They went across to the exit doors, slowly, because of their elderly mother. On the way they stopped to assist a woman to remove a baby from a high chair. They reached the exit stair but became enveloped in smoke on the stairs. Two young mothers were in the restaurant with a baby and a toddler. One woman took her friend's toddler to the toilets, which were in a deadend situation, the other remaining with the baby in a high chair. When the ®re started, she released the baby (with help) and started to move towards the exit doors. As she neared the doors she saw her friend coming from the toilets with her son. They left but were caught on the stairs by the smoke. The woman had to sit down on the stairs holding the baby and feel her way down until she emerged into clearer air at a lower level. A retired couple had also split up in the restaurant when the wife went to the toilet. When the ®re started her husband went towards the toilets with a female sta member. They both went into the female toilets and knocked on the stall doors. Several women emerged, none of whom was his wife. The man returned to the sales ¯oor only to be trapped in dense smoke at the queue that had formed at the exit doors. A member of sta (who had descended from the sta ¯oor above) entered the ®re ¯oor from these doors and directed people over towards the windows. The man managed to get out of a window onto the ledge from which he was rescued. Another elderly couple were found dead (from smoke inhalation and carbon monoxide poisoning) after the ®re in the servery area. The wife had bad arthritis. A particular problem with this ®re was the very rapid ®re growth and spread in the clothing on the wall racks. The material, mainly polyester, produced large amounts of smoke and it is estimated that the whole sales ¯oor became smokelogged within a very short time, possibly approximately 2 min after ignition. The ®re was discovered almost at ignition and the alarm raised within approximately 1 min, but this left little time for the restaurant occupants to respond and evacuate. The fact that only one set of doors was eectively available to them had an important impact on the evacuation time, a queue forming at these doors. 164 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 3.1.3. Football stadium ®re The Bradford Football stadium ®re (West Yorkshire Fire Service, 1985; Woolley, 1985; Purser, 1998) which was captured on television, provides another example of occupant behaviour in a developing emergency. As with the clothing store ®re, the video record shows occupants being rather slow to respond to the initial situation. The ®re started in accumulated rubbish under the wooden stand. During the early stages, when smoke and heat was coming up through the ¯oor, spectators cleared the immediate area but showed little sense of urgency to evacuate the stand. A number of ocials, including police were in the vicinity of the aected area, observing the smoke coming up through the ¯oor. Stand occupants remote from the seat of the ®re were initially unaected. The ®re then ¯ashed over in the space under the stand, and very soon afterwards the stand itself had ¯ashed over, so that a number of people were badly burned as they tried to escape onto the pitch. 3.2. Key aspects of occupant behaviour These few examples illustrate some of the key features of occupant response to ®re emergencies. People are not in buildings with ®re uppermost in their minds, but for other purposes to which they have a considerable commitment. In order to attend to some other demand they have to recognise its importance and cease their normal activities. The early cues to a developing ®re emergency are often ambiguous, so that some time may elapse before occupants become fully aware and convinced of the need to change their behaviour. The early cues in the clothing stores may have been ambiguous to some occupants, such as a distant commotion, or a sounder going o which might not have been a ®re alarm or might have been a ®re drill. At the football match the ®rst signs of smoke might not have been considered immediately threatening, either by the occupants or the authorities. The stand occupants were out in the open, and initially the ®re was a long way away from most of them. The clothing store and the football match also illustrate to some extent the ``friendly ®re syndrome'' in which people may not feel threatened initially and often misunderstand the rapid rate at which ®res, particularly ®re inside enclosures, can grow. Occupants may wait too long before attempting to evacuate and may become trapped. In experiments conducted for Building Research Establishment (Canter et al., 1980), subjects who were shown a series of timed photographs of a growing domestic ®re greatly underestimated the rate of ®re growth. The time people take to cease their normal activities and begin to respond to the emergency therefore depends on a range of variables. Once occupants have recognised the need to respond to the emergency, they cease their normal activities and begin a range of other activities. These activities often do not immediately include heading for the nearest exit. In the examples described, a variety of activities were engaged in. In the clothing store, in Littlewoods and probably also in the football stadium, the ®rst activities of the people near the ®re were to warn others by attracting the attention of sta. In the clothing store, the response of sta was to ®ght the ®re. In Littlewoods, the ®rst response of sta was to raise the alarm and assist evacuation, while other sta and a customer tried to ®ght the ®re. In the restaurant, the ®rst actions D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 165 were to seek information, and then to gather family members. Only when these important activities were completed did occupants begin to move towards the exits. Other activities people engage in are stopping machinery or securing money or other risks, investigative behaviour to ®nd the source of the ®re when an alarm is triggered and ®nding the appropriate exit route. When these various behaviours are completed occupants may decide to evacuate (depending upon the situation). When they decide to evacuate they ®rst make a choice of exit route and then start to evacuate. Often occupants will try to leave a building by the way they entered it and ignore other emergency exits or escape routes (Sime, 1994). The evacuation process may involve further elements of way®nding and choice of routes, depending upon the complexity of the enclosure and the developing situation (Proulx and Sime, 1991; Sime, 1998). When occupants are all together in a single enclosure, such as a shop or theatre, then the range of times to start an evacuation tend to be similar; when occupants are dispersed in a multienclosure building such as a hotel, then there is likely to be a wide variation in times to start evacuating (British Standards Institution, 1997; International Organisation for Standardization, 1999). In a building such as a hotel, when sounders are used as alarms, it is quite common for occupants to ignore them completely, particularly those remote from the ®re. If the escape route is blocked by smoke or ®re, then occupants have to make further choices about whether to continue or not, and if they continue, they may be aected physiologically by exposure to toxic smoke and heat (Purser, 1996). Another aspect of ®re safety and occupant behaviour of prime importance is ®re safety management. This includes the way in which occupants use the building and its systems before an emergency, the provision for warnings, occupant training and the ®re emergency plan as well as the management of an emergency when it occurs. The clothing store incident illustrates how delays in implementing an emergency evacuation and in shutting down tills resulted in a prolonged evacuation time. At Chester®eld, the response of sta and customers appears to have been appropriate, but the rapidity of development of the ®re in relation to the building features left little time for occupants to escape in safety. At Bradford, ®re safety management issues are evident in relation to accumulated rubbish, locked exits and delays in evacuating occupants. 3.3. Evacuation studies Evacuation studies have been conducted in a range of occupancy types in order to obtain quantitative data on emergency behaviours (Purser and Raggio, 1995; Purser and Bensilum, 1999). They also illustrate the importance of some features of occupant behaviour already described in relation to the ®re incident investigations. The following sections show a few examples. 3.3.1. Committee room The ®rst example illustrates a common ®nding, that pre-movement time is often the greatest part of evacuation time. This is important because most egress calculation 166 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 methods used for building design purposes concentrate on travel time and ignore pre-movement time, which can lead to serious underestimates of evacuation time. BRE monitored an unannounced evacuation of a 17-storey oce building in Chiswick. Part of this involved a study in an oce meeting room using a hidden camera (Purser and Raggio, 1995). The meeting room was on the third ¯oor of the building. Outside the meeting room was a foyer area, with an escape stair approximately 6 m from the meeting room door. Fig. 2 shows the position of occupants in the meeting room when the alarm sounded. The camera was in the bottom left corner of the room pointing towards the door. The occupants, who were not regular visitors to the building, were engaged in a committee meeting and were exposed to a recorded voice alarm message following a sounder alert (total alarm time 17 s). They were unaware that an evacuation was to take place. Table 2 details the recognition and response times of the each of the occupants. The occupants responded quickly after hearing the message, but engaged in activities such as gathering coats and belongings before leaving the room. One occupant left the room and then came back to retrieve a jacket. These activities required 51 s on average, while the time taken to reach the door was only 5 s. Once outside the room, approximately a further 5 s were required to reach a protected escape route. This shows a rapid and ecient response obtained using a voice alarm system. This compares with the slow response in the clothing store. 3.3.2. Class room study In further experiments subjects at Middlesex University were invited to take part in a psychological pro®le questionnaire study, which provided a cover activity for the true study (Raggio, 1996), which was to examine the behavioural responses to the alarms and compare the eciency of an alarm bell and recorded messages when motivating groups of people to evacuate. In this study, subjects signed up for one of seven dierent groups (group size 6±9 persons) which were examined at dierent times of the test day. They sat at desks in a ®rst ¯oor teaching class room covered by Fig. 2. Positions of occupants in committee meeting room during hidden camera monitored evacuation. Reproduced by permission of BRE Ltd. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 167 Table 2 Response times of occupant committee meeting room during hidden camera monitored evacuation. Reproduced by permission of BRE Ltd Person Recognition time (s) Reponse time (s) Travel to exit time (s) Total time (s) 1 2 3 4 5 6 7 8 9 10 11 12 16 15 17 20 16 18 ? 16 ? ? ? ? 40 17 20 30 30 39 ? 30 ? ? ? ? 5 5 2 3 6 5 3 13 ? ? 2 4 61 37 39 53 52 62 67 59a 55 65 71 51 Average 17 29 5 56 a Re-enters to collect jacket. Leaves again at 00:01:24. hidden video cameras and were asked to ®ll in a questionnaire. As each group settled to their task an alarm consisting of either a sounder, short voice alarm message or a longer voice alarm message was presented. Each alarm was tested on two groups (with a spare group in case a test failed). The behaviour and response times were recorded. The results (summarised in Table 3) con®rmed those from earlier studies (Bellamy and Geter, 1991; Proulx and Sime, 1991), which were that a voice alarm was more eective and produced a quicker response than a sounder. In these experiments the short message produced the quickest response. However, in these group situations it was found that the behaviour, and in particular response time, was very dependent upon a few key group members, who either facilitated or inhibited the evacuation in dierent tests. 3.3.3. Shopping centre study including a restaurant Another evacuation study in a dierent setting (Purser and Raggio, 1995; Purser and Bensilum, 1999) also shows the importance of pre-movement time and the eect of a voice alarm system. This was a study of an unannounced evacuation of Harlow shopping centre at around 10:00 hours. Included in the study was a restaurant Table 3 Pre-movement times for alarm sounder and voice message experiments. Reproduced by permission of BRE Ltd Condition n Mean (s) Standard deviation Minimum (s) Maximum (s) Sounder Long message Short message 17 15 16 35.4 22.9 12.9 15.8 4.0 7.3 16 17 5 63 31 26 168 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Fig. 3. Evacuation times for each customer leaving a restaurant. Reproduced by permission of BRE Ltd. within the shopping centre. Approximately 16 people were present including the sta. Some customers were seated while others were at the counter. Fig. 3 shows the recognition, response and movement times of 11 customers. From Fig. 3 it can be seen that the evacuation was rapid, clearance being achieved within approximately 1 min. By far the majority of the evacuation time consists of pre-movement, i.e. recognition and response times. The voice message announcement ended after 45 s, so that the movement phase either anticipated the end of the message or started within 15 s of the completion of the message. There was no obvious instruction or guidance from the sta at this stage. One occupant was a wheelchair user, whose helper reacted quickly (total pre-movement time 50 s). An elderly man near the side of the restaurant was being directed by sta 86 s after the alarm. Several plates of un®nished food were left. Once occupants have left the restaurant they still have to leave the shopping centre, although the large open mall area outside the restaurant would constitute a place of relative safety. These results illustrate that it is possible to achieve rapid clearance of a restaurant in a well managed situation. Following the Woolworths ®re it was suggested that occupants were slow (or refused) to leave food that they had paid for and that this was a factor in their being trapped by the ®re (Home Oce Fire Department, 1980; Sime, 1994). There were also suggestions that people in the restaurant were slow to move in the Littlewoods ®re (which had a sounder rather than a voice alarm; Derbyshire Fire and Rescue, 1993; Purser, 1998). However, in the Littlewoods ®re the ®re growth was very rapid and it is estimated that the restaurant occupants must have moved within a minute of the alarm sounding, so that the premovement times must actually have been relatively short. This may have been because they could see the ®re. At Woolworths the occupants in the restaurant could not see the ®re during its early stages of slow growth, which may have delayed their initial responses. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 169 3.3.4. Supermarkets and foodhalls in retail stores Monitored evacuations from two supermarkets (one attached to the Harlow shopping centre) and two foodhalls in retail stores have been studied as part of this work (Purser and Raggio, 1995, Purser and Bensilum, 1999). The two supermarkets had voice alarms systems and the foodhalls sounders. An important consideration with many shopping and assembly buildings is that two stage alarms systems are often used. The initial alarm alerts security or management, who are then able to investigate the incident before deciding whether or not to trigger the general alarm. This prevents evacuations from occurring for trivial reasons but does add a certain amount to the escape time from ignition. During real ®re incidents there are therefore two pre-movement times to consider, the pre-movement associated with the pre-alarm and the pre-movement time associated with the general alarm. The monitored unannounced evacuations for the two supermarkets started with a general alarm so the pre-alarm times were not studied, and an estimate of these times would need to be added to provide an estimate of the total evacuation time from detection of a ®re. In general the pre-movement and travel times for the supermarkets were short due to a rapid response by customers aided by sta actions. The maximum pre-movement times and times to ®nal exit are shown in a summary table, Table 4. Also summarized in Table 4 are the maximum pre-movement times and times to ®nal exit for a basement food hall in a retail store containing a food hall and a clothing section. This work was carried out in conjunction with University of Ulster and is described in more detail with their analysis of the data elsewhere (Shields et al., 1997, 2000; Ashe and Shields, 2000). The foodhall with clothing section was a single storey building with 10 exits available around the building periphery. Fig. 4 shows the results of data analysed at BRE for the distribution of pre-movement times for customers in the store. This shows a considerable variation from the ®rst people to respond (within 4 s) to the last after 1 min 50 s. The mode response time was approximately 20 s. The means of alerting the occupants was via a sounder. The video recording shows that the sta reacted within seconds to the alarm, shutting down tills and directing customers to exits (especially when a supervisor was present), except in the customer services area, where customers were served for 40 s after the alarm. In contrast, the shoppers took very little notice of the alarm sounder, the majority continuing to shop, pushing their trolleys around and taking items from the shelves. The sta responded quickly and ushered the shoppers to the exits. This emphasises the ineectiveness of sounders in buildings where the occupants are not trained to respond, and the eectiveness of a well-trained sta with an ecient emergency response plan. In general supermarkets empty quickly because a large number of sta are usually available to put into practice an emergency plan consisting of a sweep of the sales ¯oor. Fig. 5 shows the distribution of times for the total evacuation of the store (pre-movement plus movement time). This also shows a distribution of times, the ®rst person leaving immediately, the last after 2 min 45 s and a mode time of approximately 1 min. Notice that most of these distributions are skewed and the time the last person will leave (usually sta) is dicult to predict. 170 Table 4 Maximum times (decimal minutes) for phases of evacuations from experiments in dierent building types. Reproduced by permission of BRE Ltd Alarm type Recognition Response Total Travel to Total to Total to pre-movement protected route protected route ®nal exit University teaching lab Sounder Voice (long) Voice (short) 1.05 0.52 0.43 0.40 0.18 0.17 1.23 0.60 0.55 Recorded voice 0.33 0.20 0.03 0.67 0.40 0.13 1.00 0.60 0.17 0.9 0.43 0.2 0.13 ± ± ± 16 storey oce Room 311, visitors Room 312, sta Snack bar Room 208 Floor 15, staa Floor 16, staa Floor 17 staa Main stair A Main stair B Main stair D Oce building: Equinox test, visitors BRE oces/labs Building 4 Building 18 Building 19 Council oces, Amber Valley 10-storey oce, City of London Bell Voice Sounder Sounder Sounder Sounder Sounder 0.35 0.40 0.68 0.28 0.32 0.32 0.32 1.23 0.80 0.75 1.18 3.43 2.80 3.13 1.73 1.30 0.92 1.45 7.23 6.73 7.20 8.73 8.97 1.33 11.0 0.50 1.8 approx. 1.5 approx. 1.5 approx. 2.0 approx. <2 approx. 2.25 1.92 1.92 2.70 6.50 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Building Voice/announcement (End of show) (End of pres.) 3.92 2.0 approx. 1.0 approx. 6.40 4.67 1.48 Foodhall, basement Foodhall, ground ¯oor Supermarket, Herts Supermarket, Shopping centre Clothing store, Scotland Sounder Sounder Sounder? Voice ? 1.58 0.92 1.5 approx. 2.5 approx. 4.0 approx. 4.30 2.75 1.92 3.00 4.80 Hospital outpatients Waiting room 1 Eye clinic waiting room Voice 0.48 0.48 1.48 approx. 1.48 approx. Leisure centre Sounder? Library Sounder Underground station, Tyne and Wear Bell Voice 2.67 1.8 approx. 9.00 1.15 Restaurant, shopping centre 1.43 Voice 4.20 2.21 >15 5.45 4.62 a Indicates phased evacuation : pre-movement components are from ``evacuate message'', times to protected route and ®nal exit are from start of alarm. Negative pre-movement time indicates movement after ``. . .wait for further instructions. . .'' message but before ``. . .evacuate now. . .'' message. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Arts theatre Manchester theatre BRE lecture theatre 171 172 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Fig. 4. Frequency distribution of pre-movement times Ð retail store with foodhall. Reproduced by permission of BRE Ltd. Fig. 5. Frequency distribution of people leaving from all exits Ð retail store with foodhall. Reproduced by permission of BRE Ltd. 3.3.5. Theatre evacuation An example of a monitored evacuation dominated by travel time was carried out in a theatre (Purser and Raggio, 1995; Purser and Bensilum, 1999). Fig. 6 shows a plan of the ground ¯oor of a theatre including the stalls. There was also a ®rst ¯oor circle. An unannounced evacuation was carried out towards the end of an evening performance. There were approximately 160 people in the stalls and 140 in the circle, plus approximately 10 sta. This represented 46% of the full seating capacity. One exit from the stalls and the circle were considered blocked by the ``®re''. A ``double knock'' (i.e. two stage) alarm system was used and 56 s elapsed before the theatre manager came onto the stage, interrupted the performance and instructed the audience to leave, stating that the front door left and circle left exits were unavailable. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 173 Fig. 6. Theatre evacuation: ground ¯oor plan showing stalls, escape routes and exits. Reproduced by permission of BRE Ltd. Fig. 7. Theatre evacuation Ð pre-movement times in circle and stalls. Reproduced by permission of BRE Ltd. 174 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Fig. 7 shows the pre-movement times for the majority of people in the circle and stalls in camera view, timed from the beginning of the manager's announcement. The ®rst members of the audience reacted to the announcement when the manager had almost ®nished the instructions, especially in the circle. In the stalls, some people remained seated when they saw the queues formed at the exits. Although this somewhat increased the pre-movement times it had no eect on the total time required to evacuate the stalls. Fig. 8 shows the ¯ow rates at dierent doors during the evacuation time from the activation of the alarm panel. The ®rst of 300 occupants began to move after a premovement time of 12 s after the start of the manager's emergency evacuation announcement, which is 1 min 6 s after the pre-alarm was triggered. The ®rst occupants of the stall and circle emerged from the auditorium exits 30 s later, or 42 s after the beginning of the emergency announcement from the stage. There was one exit available from the circle onto two alternative stairs and three exits available from the stalls, all into the ground ¯oor foyer. In practice only one stair was used from the circle and almost all stalls occupants were encouraged to use door B. Occupants were also encouraged by sta to leave the theatre via emergency exit door C. This led to considerable congestion in the foyer and a slow total evacuation time of almost 7 min, despite the fact that the theatre was only around half full. 4. Discussion 4.1. Distributions of pre-movement and travel times A general problem with representing the data from evacuation studies is that each individual occupant has a pre-movement and travel time. When dealing with a population it is possible to consider the distributions of pre-movement times and Fig. 8. Theatre evacuation: occupant ¯ows from exits and stairs indicated in Fig. 6. Reproduced by permission of BRE Ltd. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 175 travel times, and how these interact to produce total evacuation times. When evacuation times are modelled for ®re safety engineering purposes dierent methods can be used depending upon the sophistication of the analysis modelling method. For computer evacuation simulation methods in which each occupant is modelled as an individual, it is possible to impose a pre-movement time distribution similar to those obtained in evacuation studies. For simpler evacuation calculation methods a single ®gure may be used to represent the pre-movement time of a population. Although this may be acceptable, provided the ®gure used is suciently conservative, it is not obvious which ®gure should be used. For example, it is possible to choose shortest, mean (or median) and longest times, and to defend dierent choices depending upon the situation under consideration. It may be possible to describe an evacuation quantitatively in terms of a mean and standard distribution, except that most distributions are skewed, so that some transform may be needed. Pre-movement and evacuation time distributions tend to be skewed such that some individuals take much longer to move or evacuate than the majority. Fig. 9 shows the distribution of pre-movement times for the retail store with food hall compared to a theoretical log normal distribution of times based upon the mean and standard deviation of a log transform of the data. The ®gure shows a good ®t between the observed and theoretical distributions. Several other data sets have been examined which also appear to show a log normal distribution, although this may not always be the case. In the theatre example the pre-movement times were in¯uenced by occupants observing the queues at the exits, and moving only when the queue was reduced. This resulted in a more varied distribution pattern of pre-movement times. In general it is considered that a skewed distribution such as a log normal distribution is likely to occur. It is dicult to predict when the Fig. 9. Distribution of pre-movement times in retail store with food hall compared with a log normal distribution. Reproduced by permission of BRE Ltd. 176 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 last individual will move or evacuate, since this is dependent on the behaviour of that particular individual. It may be quite possible for the vast majority of occupants to evacuate within a few minutes, but for the last occupant to emerge 20 min later. The time when the last occupant evacuates may also be an unrealistic parameter because it may vary according to the situation. In an evacuation drill for example, the last occupant to leave is likely to be a safety ocer. Since this individual would know that there was no ®re there may be no particular urgency to leave. On the other hand it is arguable that pre-movement times may be shorter in real ®re emergencies if the occupants can see the ®re. For any frequency distribution the times tend to in®nity as the frequency tends to zero, so that the time when the last individual evacuates is inherently unpredictable. It may therefore be more appropriate to describe the evacuation time for an enclosure or building in terms of a certain percentile or occupants evacuated. For example, the time to 95% premovement or clearance is reasonably amenable to calculation predictions. For most evacuation drills the time logged is the time for the last occupant to leave. Reporting the pre-movement and evacuation times of the last occupant provides a conservative and therefore ``safer'' estimate of evacuation times for engineering design purposes. Another parameter to be considered is the distribution of pre-movement and evacuation times between dierent evacuations. One consideration is the distribution of times likely to occur for dierent evacuations in the same building and how predictive a single evacuation drill is likely to be of the evacuation times for the building. Obviously evacuation times may vary according to the scenario in a real ®re incident and this is dicult to measure experimentally, although it may be amenable to modelling. It is possible to determine the variation between dierent test evacuations. It has been found that the ®re safety management plan and its implementation on any particular occasion can have a considerable eect on evacuation times. If unannounced evacuations are carried out on a number of occasions Fig. 10. Distribution of total evacuation times Ð BRE buildings. Reproduced by permission of BRE Ltd. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 177 for the same building it is possible that considerable variations might occur depending upon which sta were present on any particular occasion and their attitudes and training. Fig. 10 shows total evacuation times for 59 evacuations carried out by safety ocers of the various oce, library and laboratory buildings on the BRE site over 5 years (Purser and Bensilum, 1999). Although this set involves a number of buildings it does provide a large number of repeated evacuations in the same set of buildings under the same generic management with similar occupant populations. Apart from the common management and training situation there are a number of other similarities between the conditions for the dierent evacuations in that the occupants are similar and familiar with the buildings, the population densities are small so that queuing at exits is unlikely and the buildings are low rise (maximum four storeys) and of generally similar size. On this basis it is possible to regard conditions of the evacuation as generally similar and compare them as a single data set. On the whole the evacuations were very ecient with a narrow distribution of times. Based on an average travel time of around 0.5 min, this gives an average maximum pre-movement time of around 1 min. Fig. 10 shows the frequency distribution of times, compared with theoretical normal and log normal curves based upon the same means and standard deviations as the real data. For this set the data appear to be more or less normally distributed, as might be expected, with a mean of 1.51 min and a standard deviation of 0.55 min. 4.2. Pre-movement and travel times for a variety of buildings The examples presented in this paper have illustrated the important determinants of pre-movement time and its relationship to travel and evacuation time for a number of dierent building types with dierent management and systems. The previous section has also illustrated the variations occurring within one building set on different occasions. The work carried out by BRE has involved a wide variety of building types with a wide range of systems under a variety of managements. Some of these have been investigated in considerable detail while for others the data are only partial. However, from this data set is possible to obtain an initial set of generic data on the similarities and dierences occurring over a range of cases. The results for the set of evacuations studied are summarized in Table 4, which includes data from the examples already discussed, other examples from the BRE data set and data from the Tyne and Wear underground study (Proulx and Sime, 1991) and the Bellamy and Geter (1991) study. This gives an indication of a range of pre-movement and evacuation times occurring in a variety of buildings. One aspect all these buildings have in common is that none of them involve sleeping accommodation. They are all work places or public buildings in which the occupants would be expected to be alert. Some cases involve occupants distributed in small numbers in several enclosures (as in oce buildings), others involve occupants distributed within a single enclosure (as in shops or supermarkets) and others involve assembly spaces (such as theatres or class rooms). For some examples occupants would be expected to be familiar with the building, but less so for others, and a variety of dierent alarm systems and management are involved. 178 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 The main aspects that need to be considered are the pre-movement times and the times to evacuate into a protected escape route. In most of the cases summarized in Table 4 the total evacuation times to a ®nal exit involved exits opening directly from the occupied spaces under investigation, or the buildings were relatively small, so that time required to traverse protected escape routes was small. Exceptions were one example involving a 16 storey oce building and another a 10 storey oce building. For these cases time to ®nal exit is considerably in¯uenced by the travel times in protected stairs. However, for the 16 storey building evacuation times for occupants to enter protected escape routes were also measured. The times shown in the table are maximum times, which tend to represent the slowest responders, for both pre-movement and total evacuation times for the evacuation studies described in this report and the other published data. The data in Table 4 represent maximum pre-movement and evacuation times for a range of dierent building types, with a range of systems under a variety of managements. The results show a variety of pre-movement and evacuation times. Some times are short and similar to the BRE data (Purser and Bensilum, 1999), others show considerably longer times, particularly in situations were warning systems and ®re safety management implementation were poor. The data are plotted in Fig. 11. Unlike the BRE data, these show a better ®t with a log normal distribution. For the ®t shown, two outliers with times of 9 and 11 min were not included in the data used for the theoretical plots. These results illustrate that when systems and management are of a high standard, short pre-movement and evacuation times can be achieved fairly reliably, with similar times, in a variety of occupancy and building types. When systems and management are poor, much longer and less predictable times are obtained. It is possible that these data could be used to produce de®ned pre-movement and evacuation times for dierent de®ned levels of building system and ®re safety management. In the mixed data set the mean pre-movement time was 0.89 min and the 95 Fig. 11. Mixed occupancies Ð distribution of pre-movement times. Reproduced by permission of BRE Ltd. D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 179 percentile was 4.28 min. When this is broken down into systems using voice alarms and systems using sounders, the ®gures for voice alarm systems: mean 0.51 min, 95 percentile 2.43 min, and for sounder systems: mean 1.63 min, 95 percentile 2.57 min. The pre-movement times for evacuations in buildings using voice alarm systems were therefore generally shorter than those in buildings using sounders. To some extent this also tended to re¯ect buildings with good and less good ®re safety management. For the buildings with sounders, the dierences are even greater when the two very long pre-movement times from Table 4 (of 9 and 11 min) are included, with a mean of 2.15 min, and a 95 percentile of 7.11 min. These ®ndings support the concept that for well designed safety systems with good ®re safety management premovement times are short and predictable, while for less sophisticated and less well managed systems, pre-movement times are longer and less predictable. For the BRE data set, although the systems in most buildings are simple sounders, the presence of trained sta and a good ®re safety management environment resulted in a mean evacuation time of 1.51 min (pre-movement time <1 min) with the 95 percentile over 59 evacuations of 2.41 min. 4.3. Design considerations for pre-movement times These results indicate that for well managed oce, shops, waiting rooms and assembly spaces short pre-movement times from a general alarm can be obtained for groups of occupants. In most cases these are less than 2 min, but are subject to some variation when repeated on dierent occasions, extending up to around 4 min. One important consideration is that the majority of these data are from unannounced evacuation studies. This means that the majority of occupants, including the sta were unaware that an evacuation was about to take place. However, with any such study it is necessary to obtain the co-operation of building management at some level, and based upon our ®ndings we would suggest that, since evacuation management is so crucial in obtaining a rapid response, any knowledge of the situation by management may have some in¯uence on the outcome. Also, in general, the buildings studied tended to be well managed and the sta well trained, since such buildings and management are more likely to agree to participate in research. The long pre-movement time in the clothing store illustrates the poor performance that can occur when sta fail to conduct a proper evacuation, and we have anecdotal evidence for at least two occasions when inecient unplanned evacuations occurred (in a supermarket and a store) under the same management as those performing well in our study. Another important caveat is that none of the buildings studied involved sleeping accommodation. Based upon investigations of ®re incidents and logistical considerations it is suggested that very much longer pre-movement times would be expected in accommodation such as hotels, hostels or domestic dwellings. There is a considerable need to extend these studies so that an extensive database of pre-movement times, travel times and total evacuation times can be built up for the entire range of building and occupancy types. The best way to achieve this and obtain the most relevant data is by means of security video placed at strategic 180 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 locations in a large number of buildings, and to encourage building managers to release video records of real incidents as they occur (including both false alarms and actual ®res) for research. This could then be used to build up an anonymous database of evacuation times which could be used for design and regulation purposes. 4.4. Relationship between pre-movement time, travel time and evacuation times Another consideration when using a single number to represent pre-movement time in engineering calculations is the interaction between pre-movement time, travel time and total evacuation time (or total time to enter protected escape routes). For sparsely occupied spaces, particularly in multi-enclosure occupancies where the occupants are widely distributed (such as hotels or apartment blocks) pre-movement time is the major determinant of evacuation time, especially the pre-movement times of the slowest occupants to respond. Travel distance also becomes relatively more important. In crowded situations, such as busy shops or theatres, evacuation time depends mainly upon the pre-movement times of the ®rst occupants to react and the travel times of the whole occupant population. The ®rst occupants to begin the movement phase form the fronts of the queues at the exits and determine the time from alarm to queue formation. After this, time to clear an occupied enclosure depends mainly upon the exit choice behaviour and the exit route ¯ow capacities. A good example of this phenomenon is the theatre evacuation. In this case many occupants remained seated until they could see the queues at the exits shorten (the sort of behaviour common in airport lounges after a ¯ight is called). In such a case, simply adding the maximum pre-movement time to the total travel time for the occupant population would give an unrealistically long estimate of total evacuation time. In other examples illustrated, such as the restaurant evacuation, the occupant capacity was insucient to result in queuing at the exits, so that the pre-movement time of the slowest responders and their travel times can be summed to provide a reasonable estimate of total evacuation time. For the retail store with food hall an intermediate situation occurred, but since the exits never reached their maximum ¯ow capacity there was little evidence of queuing and both pre-movement and travel times were important in terms of total evacuation time. 5. Conclusions 1. Although the detailed behaviour and emergency evacuation times of individual building occupants may be somewhat unpredictable, the behaviour and evacuation times of occupant groups and building populations are amenable to prediction and quantitative description suitable for engineering design purposes. 2. Evacuation behaviours and times can be described in terms of a limited set of design behavioural scenarios (related principally to the standard of ®re safety management and warnings provision, the size of the occupant population within enclosures in relation to exit capacity, the distribution of occupants within and among building enclosures and whether occupants are alert of sleeping). D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 181 3. In buildings with a good standard of ®re safety management and ®re safety systems, evacuation times are short and predictable within narrow limits. When standards of ®re safety management and systems are poor, evacuation times become much more variable. 4. Evacuation times depend upon two major elements, pre-movement times and movement (or travel) times. 5. Movement times are reasonably predictable according to simple established ¯ow calculation methods, providing exit choice is correctly predicted. Movement times can be considerably lengthened (particularly in assembly enclosures) when occupants favour some available exits over others. 6. Pre-movement times are subject to a wide range of in¯uences but tend to be short and predictable in well managed buildings with good ®re safety systems (with the principal exception of multi-enclosure sleeping accommodation). 7. For buildings involving a small range of evacuation classes in terms of dierent occupancy types and design behavioural scenarios, it should be possible to de®ne default pre-movement times or time distributions which could be used with simple egress calculation methods to calculate design evacuation times for ®re safety engineering. For larger and more complex buildings involving more varied occupancies then more sophisticated computer evacuation models may be required. 8. An extensive database of pre-movement times, travel times and total evacuation times should be built up for the entire range of building and occupancy types by encouraging building managers to release video records of real incidents as they occur (including both false alarms and actual ®res) for research. This could then be used to build up an anonymous database of evacuation times which could be used for design and regulation purposes. References Ashe, R., Shields, T.J., 2000. Analysis and modelling of the unannounced evacuation of a large retail store. Fire and Materials 23, 333±336. Bellamy, L.L., Geter, T.A.W, 1991. Experimental programme to investigate informative ®re warning characteristics for motivating fast evacuation. Building Research Establishment Report BR 172. Building Research Establishment Ltd, Garston, Watford, UK. British Standards Institution, 1997. BS DD249 Fire safety engineering in buildings. Part 1. Guide to the application of ®re safety engineering principles. British Standards Institution, London. Canter, D., Breaux, J., Sime, J., 1980. Domestic, multiple occupancy and hospital ®res. In: Canter, D. (Ed.), Fires and Human Behaviour. Wiley, Chichester (Chapter 8) pp. 117±136. Derbyshire Fire and Rescue Service, 1993. Fire investigation report. Littlewoods Store, Market Place, Chester®eld Ð 7th May 1993. Derbyshire Fire and Rescue Service. Home Oce Fire Department, 1980. Report of the planning/legislation sub-committee on the ®re at Woolworth's, Piccadilly, Manchester on 8 May 1979. Joint ®re prevention committee of the central ®re brigades advisory councils for England and Wales and for Scotland. Home Oce Fire Department, London SW1. International Organsiation for Standardization, 1999. Fire safety engineering Ð part 8: life safety Ð occupant behaviour, location and condition, ISO/TR13387-8. International Organization for Standarization, Geneva 182 D.A. Purser, M. Bensilum / Safety Science 38 (2001) 157±182 Nelson, H.E., MacLennan, H.A., 1995. Emergency movement. In: DiNenno, P.J. (Ed.), SFPE Handbook of Fire Protection Engineering, National Fire Protection Association, Quincy, MA (Section 3/Chapter 14) pp. 286±295. Pauls, J.L., 1995. Movement of people. In: DiNenno, P.J. (Ed.), SFPE Handbook of Fire Protection Engineering, National Fire Protection Association, Quincy, MA (Section 3/Chapter 13), pp. 263±285. Proulx, G., Sime, J.D., 1991. To prevent panic in an underground emergency, why not tell people the truth? In: Cox, G., Langford, B. (Eds.), Fire Safety Science Ð Proceedings of the Third International Symposium. Elsevier Applied Science, New York, pp. 843±852. Purser, D.A., 1994. Hazard perception and behavioural response of people to smoke exposure in accidental ®res Ð report on preliminary case investigations. Building Research Establishment Ltd, Garston, Watford, UK. Purser, D.A., 1996. Behavioural impairment in smoke environments. Toxicology 115, 25±40. Purser D.A., 1998. Quanti®cation of behaviour for engineering design standards and escape time calculations. Institute of Building Control Ð College 98. Brighton. 19±22 March 1998. Building Research Establishment Ltd, Garston, Watford, UK. Purser, D.A., Raggio, A.J.T., 1995. Behaviour of crowds when subjected to ®re intelligence. Building Research Establishment Report CR 143/95. Building Research Establishment Ltd, Garston, Watford, UK. Purser, D.A., Bensilum, M., 1999. Quanti®cation of escape behaviour during experimental evacuations. Building Research Establishment Report CR 30/99. Building Research Establishment Ltd, Garston, Watford, UK. Purser, D.A., Kuipers, M., Bensilum, M., 1998. Buildings, ®re and occupant behaviour Ð a relational database. Building Research Estabishment Report CR 386/98m. Building Research Establishment Ltd, Garston, Watford, UK. Raggio, A., 1996. Comparative eciency of alarm bell and recorded messages when motivating groups of people to evacuate. Building Research Establishment Report CR 147/96. Building Research Establishment Ltd, Garston, Watford, UK. Shields, T.J., Boyce, K.E., Silcock, G.W.H., 1997. Unannounced evacuation of Marks and Spencer Spruce®eld store, Belfast. Report to Department of the Environment, London. Shields, T.J., Boyce, K.E., Silcock, G.W.H., 2000. Towards the characterization of large retail stores. Fire and Materials 23, 325±331. Sime, J.D., 1994. Human behaviour in domestic and hospital ®res. Building Research Establishments Occasional Paper OP59. Building Research Establishment Ltd, Garston, Watford, UK. Sime, J.D., 1998. An occupant response escape time (ORET) model. Human Behaviour in Fire Ð Proceedings of the ®rst international symposium. University of Ulster, pp. 299±308. Carrickfergus, Co. Antrim, Northern Ireland, UK. West Yorkshire Fire Service, 1985. Fire research report Ð ®re incident at Bradford City AF Club Ltd, Valley Parade Ð 11 May 1985, Birkenshaw. West Yorkshire Fire Service, Birkenshaw, UK. Woolley, W.D., 1995. A scientist's view of the ®re at Bradford City football stadium. Notes from Research Colloquium `Fire and the Environment' at the Fire Research Station, Borehamwood, 6 November 1985 (Notes accompanying video of ®re incident). Building Research Establishment Ltd, Garston, Watford, UK.