Estimating behavioral changes for transportation modes after terrorist
attacks in London, Madrid, and Tokyo
Final Report
METRANS Project 08- 10
March 2010
Detlof von Winterfeldt
Fynnwin Prager
National Center for Risk and Economic Analysis of Terrorism Events ( CREATE)
University of Southern California
Los Angeles, 90089- 2902
Disclaimer
The contents of this report reflect the views of the authors, who are responsible for the
facts and the accuracy of the information presented herein. This document is
disseminated under the sponsorship of the Department of Transportation, University
Transportation Centers Program, and California Department of Transportation in the
interest of information exchange. The U. S. Government and California Department of
Transportation assume no liability for the contents or use thereof. The contents do not
necessarily reflect the official views or policies of the State of California or the
Department of Transportation. This report does not constitute a standard, specification,
or regulation.
Abstract
Why do individuals change their behavior after terrorist attacks? To what extent do
changes in risk perception explain changes in travel behavior? This project aims to
answer these questions by examining the three major attacks in recent history on public
transit systems: the London bombings ( July 2005), the Madrid bombings ( March 11,
2004), and the Sarin Gas attacks in Tokyo ( March 20, 1995). Each case is found to be
unique. Reductions in passenger journeys on attacked transportation modes range from an
average of 10 percent over 20 weeks in London to no significant change in Tokyo, while
substitution to alternative modes also varies across cases. This variance is likely due to
more than cultural difference, with primary attack characteristics, transportation system
factors, and the social amplification of risk perceptions also playing a role. Such findings
have important implications for policy makers and academics with an interest in
transportation security and the behavioral and economic impacts of terrorist attacks.
Table of Contents
Page 8 Comparing behavioral responses to terrorist attacks on public transit systems:
London, Madrid, and Tokyo.
Fynnwin Prager† and Detlof von Winterfeldt†
Page 21 Exploring reductions in London Underground passenger journeys
following the July 2005 bombings
Fynnwin Prager†, Garrett Beeler Asay†, Detlof von Winterfeldt†, and
Bumsoo Lee†
Page 43 A study of the impact of the July bombings on Londoners’ travel behavior
Barbara Fasolo††, Zhifang Ni††, and Lawrence D. Phillips††
Page 60 The impact of the 3/ 11 Madrid bombings on consumers travel behavior
Thomas Baumert†††
Page 79 Analysis of passengers’ reactions to the sarin gas attacks in Tokyo
Fynnwin Prager†, Barbara Fasolo††, and Zhifang Ni††
† National Center for Risk and Economic Analysis of Terrorism Events ( CREATE),
University of Southern California
†† Operational Research Group and Decision Capability Unit, London School of
Economics and Political Science
††† Universidad Complutense de Madrid, Universidad Católica de Valencia “ San Vicente
Mártir”
List of facts and figures
Comparing behavioral responses to terrorist attacks on public transit systems: London,
Madrid, and Tokyo.
Fynnwin Prager† and Detlof von Winterfeldt†
Page 13 Table 1: Comparison of dread hypothesis results across cases
Exploring reductions in London Underground passenger journeys following the July
2005 bombings
Fynnwin Prager†, Garrett Beeler Asay†, Detlof von Winterfeldt†, and Bumsoo Lee†
Page 35 Table I: Variables included in the regression model and expected
coefficient signs
Page 36 Table II: Time Variables Equations
Page 37 Table III: Monthly change in London Underground passenger journeys
( 2005- 2006)
Page 38 Table IV: Regressions of daily London Underground passenger journeys
( 2001- 2007)
Page 39 Table V: Summary statistics for regression analyses
Page 40 Figure 1: London Underground passenger journeys, all lines, observed and
predicted ( 2003- 2006)
Page 41 Figure 2: London Underground passenger journeys, all lines, observed and
prediction 95- percent confidence intervals ( 2003- 2006)
Page 42 Figure 3: Change in London Underground aggregate weekly gate
entrances by line grouping ( 2005)
A study of the impact of the July bombings on Londoners’ travel behavior
Barbara Fasolo††, Zhifang Ni††, and Lawrence D. Phillips††
Page 46 Table 1: Interannual variations of London Underground entry: 2005 versus
2004
Page 46 Figure 1: Weekly tube usages in 2005 compared to 2004
Page 47 Figure 2: Yearly traffic volume of bus or coach in London
Page 48 Table 3: Yearly London traffic volume and interannual variations
Page 49 Table 4: Annual fatalities by transport mode
Page 49 Figure 4: Annual fatalities by transport mode
Page 50 Table 5: Interannual variations of fatality in London by mode
Page 50 Figure 5: Interannual variations of fatality in London by mode
Page 50 Figure 6: London Monthly fatalities for Pedal Cycles, Powered- two-wheelers
and Cars & Taxis.
Page 51 Table 6: Six- month fatality ratios ( Jul- Dec/ Jan- May) between 2002 and
2005
Page 53 Table 7: Key year- on- year changes in traffic entering the central London
charging zone during charging hours
Page 54 Table 8: Yearly fatality rate in persons killed per million vehicle
kilometers
Page 55 Table 9: Share of fatalities by mode
Page 56 Figure 7: Percentage share of the fatalities of the three directly- hit
boroughs to the London total
Page 57 Figure 8: Monthly casualties of Pedal Cycles, 2- wheeled motor vehicles
and Cars & Taxis in London.
The impact of the 3/ 11 Madrid bombings on consumers travel behavior
Thomas Baumert†††
Page 63 Figure 1: Daily number of passengers Madrid Bus ( 2002- 2004)
Page 65 Figure 2: Weekly distribution of Madrid bus passengers ( 2003- 2006)
Page 66 Figure 3: Weekly distribution of Madrid bus passengers ( 2003- 2006)
Page 68 Figure 4: Number of daily passengers Madrid Metro March ( 2002- 2007),
All lines
Page 68 Figure 5: Number of daily passengers Madrid Metro March ( 2002- 2007),
Line N. 1
Page 69 Figure 6: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 2
Page 69 Figure 7: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 3
Page 70 Figure 8: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 4
Page 70 Figure 9: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 5
Page 71 Figure 10: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 6
Page 71 Figure 11: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 7
Page 72 Figure 12: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 8
Page 72 Figure 13: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 9
Page 73 Figure 14: Number of daily passengers Madrid Metro March ( 2002- 2007)
Line N. 10
Page 75 Figure 15: Average interannual variation in the number of train travellers,
highway vehicles and fatal highway accidents in March and April 1999
through 2003 versus March and April 2004 in Spain
Page 78 Appendix I: Madrid Metro Train Route Map
Analysis of passengers’ reactions to the sarin gas attacks in Tokyo
Fynnwin Prager†, Barbara Fasolo††, and Zhifang Ni††
Page 82 Figure 1: Time series of monthly passenger volume, 1992 to 1998
Page 83 Figure 2: Monthly change in total passengers, 1992 to 1998 (‘ 000s)
Page 84 Table 1: Predicted change in TRTA passenger numbers 1995, January-
June
Page 84 Figure 5: ARIMA model predictions, all passengers
Page 88 Figure A1: Monthly change in Season ticket passengers, 1992 to 1998
(‘ 000s)
Page 88 Figure A2: Non- season ticket passengers (‘ 000s)
Page 89 Figure A3: ARIMA model predictions, season tickets
Page 89 Figure A4: ARIMA model predictions, non- season tickets
Page 90 Figure A5: ARIMA model prediction errors, season tickets
Page 90 Figure A6: ARIMA model prediction errors, non- season tickets
Disclosure
Project was funded in entirety under this contract to California Department of
Transportation.
Comparing behavioral responses to terrorist attacks on public
transit systems: London, Madrid, and Tokyo
Fynnwin Prager* and Detlof von Winterfeldt
National Center for Risk and Economic Analysis of Terrorism Events ( CREATE)
University of Southern California
* fprager@ usc. edu
Abstract
This section introduces, summarizes, and compares the four studies in this project.
Understanding behavioral responses to terrorist attacks on public transit systems is
important so that policy makers may better mitigate their economic and human costs. Yet
relatively little research exists on the subject due to the significant focus on the impacts
of 9/ 11 as well as the lack of public transit attacks on US soil. To address this
shortcoming, the studies in this project examine three major attacks on public transit
systems in London, Madrid, and Tokyo. Comparing across these studies, we find unique
responses to each attack. For example, passenger journeys on attacked transportation
modes were reduced by an average of 10 percent over a period of 20 weeks in London,
while no significant change was observed in Tokyo. We explore reasons for this variance,
suggesting that primary attack characteristics, transportation system factors, and the
social amplification of risk perceptions appear to play a role.
Introduction
Public transit systems are a common target for terrorist attacks worldwide. Their
attractiveness as a target for attack is obvious. Primarily, transit systems carry large
numbers of individuals in confined spaces, providing the opportunity for terrorists to kill
many people with low- cost weapons. Moreover, most transit modes feature low- security
vehicles that are also vulnerable – for example, they do not have shock- resistant
structures. Finally, public transit systems often sit at the heart of broader transportation
and economic networks. Disrupting a transit system can therefore cause substantial harms
to a region’s economy.
The immediate impact of terrorist attacks on public transit systems is both horrific
and well documented, yet the so- called “ secondary” impacts are less known. Public
transit systems were attacked 182 times worldwide between 1997 and 2000, with 37
percent of attacks involving fatalities, a far higher proportion than terrorist incidents in
general ( Jenkins, 2004). The human toll of these fatal attacks is significant, with 10 or
more fatalities occurring in 28 percent ( Jenkins, 2004). As evidenced by our case studies,
deadly attacks on public transit systems have occurred both before and since that period.
In contrast, neither the cost of property damage caused by these attacks, nor the
“ secondary” economic and human costs, has been documented. This lack of research
provides the impetus for this project.
There is some research into “ secondary” behavioral responses to terrorism attacks
on transportation systems more generally, yet little on public transit systems specifically.
This is unsurprising give the substantial effort to understand the impacts of the attacks on
airlines on September 11th 2001, as well as the lack of attacks on public transit systems in
the US. Yet the high incidence of attacks worldwide, along with the potential for
significant economic impact – transit systems provided over 9 billion passenger journeys
per year at the turn of the last decade ( Guerrero, 2002) – means that academics and
policymakers alike should be concerned with this issue area.
Social psychologist Gerd Geigerenzer ( 2004, 2006) provides a useful conceptual
framework – the “ dread hypothesis” – with which to understand behavioral responses to
terrorism attacks on transportation systems in general. The “ dread hypothesis” comprises
three connected stages. The first stage is “ dread avoidance,” reflected in a reduction in
passenger journeys on the attacked transportation mode. The second stage “ substitution,”
is shown by an increased use of alternative, non- attacked transportation modes. There is a
critical assumption underlying these first two stages: The reduction in passenger journeys
is the result of demand side changes in risk perception, and not due to other influential
demand ( economic wealth, prices, e. g.) or supply ( service provision, congestion)
variables for attacked and substitution modes alike. The dynamics of this process – that is
the interactions of these other supply and demand variables – is also important to
consider. For example, reductions due to fear may be underestimated if they are offset by
an uptake in ridership due to reduced congestion. The third stage is an increase in
fatalities which result from this substitution, implying that the substitution transportation
mode has higher fatality rates than the attacked mode.
In terms of the first stage, numerous studies have identified reductions in
passenger journeys following terrorist attacks. In recent years there has been a particular
focus on responses to the the September 11th 2001 attacks ( Gigenrenzer, 2004, 2006;
Blalock, Kadiyali, & Simon, 2005; Sivak and Flanagan, 2003; Ito and Lee, 2005; Beeler
Asay & Clemens, 2008; Gordon et al, 2007) with an estimated overall reduction in
passenger journeys of 6 percent over 2 years. In their study of Israel, Becker and
Rubinstein ( 2004) estimate that an attack is likely to reduce the number of bus passenger
journeys by around 30 percent during the 2 months following an attack, while López-
Rousseau ( 2005) obverses a reduction of 4- 6 percent for the 2 months following the 2003
Madrid attacks It is important to note that these studies employ varying degrees of
sophistication in modeling this process, as discussed further in the study by Prager and
colleagues below. Nonetheless, López- Rousseau ( 2005), reflecting on the findings in all
three cases, suggests, “ avoiding a dread risk [ the fear of an event occurring] is a universal
effect.”
In contrast, studies of the second and third stages provide conflicting results in
different contexts. Here Gigenrenzer ( 2006) and Blalock, Kadiyali, and Simon ( 2005)
find that US residents shifted transportation mode from airlines to private road vehicles.
Sivak and Flanagan ( 2003) and Gigenrenzer ( 2006) show respectively the two elements
of the third “ dread hypothesis,” that the substitute mode is more risky in terms of
fatalities, and that fatalities increase as a result of the transportation mode shift. Indeed,
10
the latter estimates that some 1,500 additional individuals died in the US as a result of
this mode shift, highlighting the potential human cost of secondary impacts. However,
unlike the US, López- Rousseau ( 2005) finds no increase in alternative modes of
transportation. In turn, there was no increase in accidents or fatalities on these other
modes following the attacks. In the Israel case, Becker and Rubinstein ( 2004) find
evidence of shifts to alternative transportation modes is found, with increases in taxi
passenger journeys in particular. However, they did not examine the fatality element of
the “ dread hypothesis” framework.
It is important to examine the psychological dimensions of this “ dread
hypothesis.” A key theoretical point in the literature on behavioral responses to risky
events such as terrorism is the focus on risk perception as opposed to the statistical
likelihood of that event. What may appear to the statistician as “ irrational,” or the
neglecting of calculable probabilities ( Sunstein, 2003), is instead individuals responding
to what they perceive to be the threat. Such risk perceptions are emotional rather than
calculated – they are subject to worry ( Sjoberg, 1998) or dread ( Fischhoff et al, 1978;
Slovic 1987) – they are dynamic rather than fixed, they are subjective rather than
objective, and most importantly, they are socially amplified ( Kasperson, Renn, Slovic,
Brown, Emel, Goble, Kasperson, & Ratick, 1988; Kasperson, 1992; Kasperson,
Kasperson, Pidgeon, and Slovic, 2003). A fully developed conceptual framework for the
social amplification of risk is presented in Kasperson et al ( 2003), which usefully
incorporates the interrelating and dynamic influences of various government and media
agencies, cultural and social norms and values, and personal social networks.
A second theoretical point is that individual behavior in response to risk
perceptions can vary. Lerner et al ( 2003) provide a comprehensive presentation of
academic literature on this subject. A key finding here is that individual perceptions of
terrorist events associated with anger are likely to be met with behavioral responses of
“ certainty and individual control,” while individual perceptions of terrorist events
associated with fear are likely to be met with behavioral responses of “ pessimistic
estimates and risk averse choices” ( Lerner et al, 2003).
Case studies: Summary of papers
Exploring reductions in London Underground passenger journeys
following the July 2005 bombings
Prager, Beeler Asay, Lee, and von Winterfeldt use a multivariate time- series
regression model to examine the impact of the London July 2005 bombings on London
Underground passenger journeys. They find an estimated reduction of 22.5 million fewer
passenger journeys over the 4 months following the attacks. Our analysis suggests that
heightened risk perceptions are a significant cause of reduced Underground travel,
accounting for around 82 percent of passenger journey reductions following the attacks.
Lines affected by the bombings appear to have experienced particularly high reductions
in passenger journeys. The data also suggests that passenger journeys following the
attacks were reduced to a greater extent at weekends and holidays compared with
11
weekdays. This is notable because the majority of travel on weekdays is for work and
education, while the majority of travel on weekends is for shopping and leisure trips.
Their estimations thus suggest an extra impact for the central London retail and tourism
economy.
Prager, Beeler Asay, Lee, and von Winterfeldt’s estimates control for both
demand ( such as demographic, economic, and weather) and supply ( station closures,
service disruption, time delays) variables. The combination of controlling for these
factors, along with the period of reduction extending beyond the reopening of stations
after repairs, suggests that changing risk perceptions played a role in the reduction of
passenger journeys following the attacks. This finding is supported by survey data
( Goodwin et al, 2005; Rubin et al, 2005, Rubin et al, 2007) which shows that 19 percent
of respondents reported traveling less as a result of the attacks. Aggregate data limits the
ability for close inspection of this issue, such as whether particular social groups were
more likely to choose not to travel by the Underground, or which transport modes
individuals switched to.
A study of the impact of the July bombings on Londoners’ travel behavior
Fasolo, Ni, and Phillips use the “ dread hypothesis” model employed by both
Gigenrenzer ( 2004, 2006) and López- Rousseau ( 2005) to study the impact of the London
July 2005 bombings on passenger behavior. They find that Londoners’ responses to the
July 2005 bombings were distinct from both US and Spain resident’s reactions to the
respective attacks. In line with US and Spain residents, Londoners appear to have
avoided attacked modes – buses as well as Underground. Like US residents, Londoners
increased their use of alternative modes, in this case pedal cycles and powered- 2-
wheelers. However, like the Spain case there is no evidence of increased fatality rates in
London.
Fasolo, Ni, and Phillips explore empirically a number of explanations for these
unique results. They rule out the suggestion that that substitute modes were less risky
than attacked mode by showing that per kilometer risk is higher for the former. They also
reject the argument that fatalities in London were focused around the area of the attacks
by examining the spatial spread of fatality rates. Moreover, they examine accident rates
and find they did not increase either.
Analysis of passengers’ reactions to the sarin gas attacks in Tokyo
Prager, Fasolo and Ni use monthly Tokyo subway passenger data to study the
impact of the March 1995 sarin gas attacks in which 12 died. They employ univariate
time series regression analysis to explore whether the first step of the “ dread hypothesis”
is correct. Though a slight reduction below predicted levels is observed, this is deemed
insufficient to reject the alternative hypothesis that no reductions in passengers journeys
was experienced following the attacks. This finding stands Tokyo in contrast with the
behavioral responses of US, Spain, and London residents, and implicitly rules out the
potential for secondary impacts relating to transportation use.
12
Prager, Fasolo, and Ni explore a number of reasons for these distinct findings. On
the one hand, the absence of a significant change in the use of the attacked mode is
explainable due to the limited transportation alternatives, the relatively low number of
deaths resulting from the attacks, especially when compared with the 6,000 plus deaths
experienced in the Kobe, Japan earthquake two months earlier, and the limited service
disruption given the lack of damage to subway infrastructure. It may also be that
reductions due to fear were offset by an uptake in ridership due to reduced congestion.
On the other hand, this distinct finding is surprising given the relatively slow response of
the Japanese government, especially in arresting and convicting culprits, subsequent
attacks, and perhaps most importantly, the unprecedented nature of the attacks.
The impact of the 3/ 11 Madrid bombings on consumers travel behavior
Baumert develops the López- Rousseau ( 2005) study of Madrilenian reactions to
the 3/ 11 bombings. He presents bus and metro passenger journey data to complement the
train and car passenger and fatality data highlighted by López- Rousseau. This is an
important development because the transportation patterns within Madrid have not
previously been examined. Baumert finds that both buses and metro operators experience
a single day drop in passenger journeys on the day of the attacks, with figures bouncing
back following the attacks. Unfortunately, passenger journey data for the short- distance
inter- urban trains directly affected by the attacks is unavailable.
Baumert suggests that, in line with López- Rousseau, the Spanish resident
responses to the 3/ 11 attacks are tempered relative to the US and London cases due to the
decades- long history of terrorism on Spanish soil. Moreover, the relatively limited size of
attack and smaller “ car culture” compared with the US mitigated transportation behavior
impacts compared with the September 11th 2001 aftermath. In terms of Madrid intra-urban
transportation substitutes data, it seems that only a short- term impact occurred,
suggesting that where passengers did move away from the inter- urban train system – as
shown by López- Rousseau – they moved neither to cars nor to the metro or bus systems.
This would suggest that passengers decided not to travel rather than choose alternative
modes.
Comparison of cases: London, Madrid, and Tokyo
The first key theoretical finding is that transportation behavioral responses to
terrorist attacks are far from uniform. In particular, the Tokyo case suggests the “ dread
risk” avoidance is not “ universal” as López- Rousseau ( 2005: 427) suggests. If true, this
raises important questions. Why have public responses to these attacks appeared to differ
between cases and countries? What variables distinguish these cases? In order to examine
the case studies in a coherent manner, it is important to develop a framework for analysis
based upon current theory on behavioral responses to terrorist attacks.
13
We suggest a tentative framework for analysis. This builds upon the binary
primary/ secondary impact model of hazardous events proposed by Kasperson ( 1992) and
the social amplification model of risk model proposed by Kasperson et al ( 1988). Primary
impacts are those direct results of the hazardous event, such as lives lost, traumas
induces, structures destroyed and infrastructures damaged. Secondary impacts are those
hazardous event impacts which “ extend beyond the people directly affected by the
original hazard event or report” ( Kasperson, 1992: 160).
Table 1: Comparison of dread hypothesis results across cases
Dread hypothesis stages
Location, Date
( mode and method
of attack)
Change in aggregate
attacked mode
passenger journeys
Change in
alternative mode use
Change in
alternative mode
fatalities and
accidents
Tokyo, Japan,
March 20th 1995
( subway, Sarin gas)
No significant
reductions ( Prager,
Fasolo, & Ni,
below)
No change ( implied) No change ( implied)
US, September 11th
2001 ( airlines,
crash into buildings)
6% average
reduction over two
years ( Gordon et al,
2007)
Increase in private
road vehicles
1,500 additional
road deaths
Madrid, Spain,
March11th 2004
( train, bombs)
5% average
reduction over two
month ( López-
Rousseau, 2005)
No significant
increase in road use
( López- Rousseau,
2005). Single day
reduction in buses
and metro ( Baumert,
below).
No change ( implied)
London, UK, July
7th 2005 ( subway
and bus, bombs)
8.3% average
reduction over 4
months ( Prager et
al, below)
Increase in pedal
cycle and two-wheeler
use ( Fasolo,
Ni, & Phillips,
below).
No significant
increase in fatality
or accident rates
across alternative
modes and localities
( Fasolo, Ni, &
Phillips, below)
14
Primary attack characteristics
The first set of variables likely to influence public transportation choices
following terrorist attacks is primary attack characteristics such as the method, size, scope
and location of the attacks, as well as the impacts of the attack, such as the number of
deaths, injuries, and the damage caused. For example, a large, coordinated attack on
numerous points in a transportation system would likely result in greater reductions in
passenger journeys for that mode when compared with a relatively minor attack.
However, this set of variables is only manifested through the following sets of variables,
the transportation system factors and social amplification factors.
Transportation system factors
First, the primary attack characteristics are filtered through transportation system
factors. These include the extent of damage relative to size of the system, the difficulty to
repair any damage, the number of points damaged, and the flexibility of the system in
terms of alternative routes. The key variable here is the cuts in service, which result from
damage done and contribute to reductions in passenger journeys for the attacked mode. A
clear distinction here is between the sarin gas attacks of Tokyo and the bombings on the
Madrid rail and London Underground systems. The lack of infrastructural damage caused
by the chemical attacks meant that the Teito Rapid Transit Authority was able to resume
service quickly following the attacks. This stands in contrast to the London case where
full service was not resumed for a month following the attacks.
Transport mode substitutability within the broader transportation system of the
urban area appears to be important also. The ability of individuals to switch to other
forms of transport to avoid the “ dread risk” of the attacked mode appears to influence the
change in passenger journeys following the attacks. For example, the lack of reduction in
passenger journeys in the Tokyo case are likely to have been influenced by the
inflexibility of the broader system to cope with alternative routes. In contrast, the relative
high flexibility in the US transportation system enabled individuals to take alternative
modes. A lack of flexibility in the broader system would make the change in passenger
numbers more reactive to the time taken to repair damage in the system. The relationship
between transportation modes is also a factor, particularly in reference to travel time and
service quality. Of course, the communications technology revolution of the past few
decades has enabled a growing flexibility in transportation choices, such as the ability to
work from home or shop online.
Social amplification
The complex nature of social amplification, as discussed above and in Kasperson
et al ( 2003), constrains precise identification of influencing variables. However, there are
some elements of social amplification which appear to have influenced the transportation
behavior change following terrorist attacks. A first general point to make is that changing
risk perception can play a role in influencing transportation choice. Evidence from the
Prager et al paper in this project suggests that this was the case following the London
15
2005 bombings, and survey data for both the UK ( Goodwin et al, 2005; Rubin et al, 2005,
Rubin et al, 2007) and US ( Schuster et al, 2001; Schlenger et al, 2002; Lerner et al, 2003)
supports this finding. However, this does not appear to have been the case following the
attacks in Tokyo.
In this Tokyo case, the role of previous events, specifically the Kobe earthquake
two months prior, may have mitigated the social amplification of the terrorist attacks.
Equally, the lack of major hazardous events prior to the other terrorist attacks researched
in this project may have heightened their shock and impact. The attacks in London,
Madrid, and the US were all unprecedented events that appear to have been met with new
reactions. Clearly, the relative impact of any attack to previous events is important here.
An important element of the social amplification of risk is that responses to
specific hazardous events are likely to be unique across sections of society. This
emphasizes the point that culture can play an influential role in transportation mode
choice following terrorist attacks both within and between nations and cultures. Indeed,
the key debate within the literature on this point is the tension between universalism and
cultural relativism, or “ cultural theory,” with evidence appearing to support both sides.
As risk perception theorists Bernd Rohrmann states:
“ A central idea in Cultural Theory is that people in their risk perceptions express
cultural biases which in turn “ support” different patterns of social relations.
Several attempts have been made to investigate how large a part of risk perception
could be explained by cultural aspects, but research on the topic shows diverging
results” ( Rohrmann, 2000: 178).
For example, Dake ( 1991) finds evidence to support cultural theory while Sjoberg ( 1997,
1998) cannot verify this position, and Brenot & Bonnefours ( 1994) and Goszczynska
( 1991) both find evidence to support the universal perspective.
In sum, we suggest that the public reaction to terrorist attacks on public
transportation systems are influenced by the primary attack characteristics as manifested
through the systemic factors and an interactive element of the secondary, socially
amplified media, government and public responses, which are clearly also contextual.
The purpose here is to explore potential universal variables, as opposed to universal
effects per se. It is important to note that not all of these variables are measured within
this set of papers.
Implementation Section
These findings have important consequences for policy makers interested the
secondary impacts of terrorist attacks. First, our study highlights the potential for
reductions in use of attacked transport modes, which have the potential to cause
subsequent economic harms and reductions in social welfare. However, these impacts are
far from uniform, with divergent results apparently the consequence of distinct primary
attack characteristics, social amplification, and transportation system factors.
16
Second, our findings suggest that supply side factors can influence passenger
reductions following the attacks. The London bombing results show that a combination
of increased station closures, increased delay times, and reduced service operation all
combined to account for around 18 percent of passenger journey reductions for the 4
months following the attacks. This proportion could have been far more substantial had
the London Underground not resumed service so quickly, with all lines in operation by
August 4, less than a month after the attacks. This highlights the importance of service
provision in minimizing the secondary impacts of terrorist attacks.
Third, the results suggest that compounding incidents – in this case the failed
attacks of July 21 2005 and the Police killing of an innocent individual – have the
potential to increase reductions in passenger journeys. This suggests that policy makers
and security officials must balance the potentially conflicting aims of halting multiple
attacks while limiting disproportionate security responses. Further research is required to
identify the factors which achieve this aim, though such approaches could include
increasing non- violent police presence and the incidence of randomized security checks.
Our findings also indicate that policy efforts following terrorist attacks should
focus on reducing public risk perception of travel on the affected mode. A key
consideration in designing appropriate policy responses is to work towards aligning risk
perceptions with risk reality. Risk communication by policy makers after the event has to
be crafted in a way that neither unnecessarily alarms nor provides false comfort to
people. Actions often speak louder than words. For example, after the London liquid
bomb scare of 2006, the US Department of Homeland Security banned all liquids from
planes. To some this appeared to be an overreaction, given statistical risk of an attack of
this type, but it also did appear to lower public fears and, as a result, had only a minor
effect on air travel.
Our findings highlight the opportunity for further research in this area. While it
appears that risk perception may play a role in the London bombings case, it is yet to be
explored whether the same results have occurred in further terrorist attacks on
transportation systems worldwide. Comparison between these cases would enable
research to examine the influence that different attack variables – such as the size, type,
and location of attacks – may have on risk perception and behavioral responses. It would
also be instructive to compare terrorist events with non- terrorist hazards and accidents as
this would allow for more general risk perception findings to be revealed, such as the rate
at which passengers return to pre- attack mode choices. All such findings have important
economic and policy making implications which have also yet to be explored fully.
17
Future research
Future research could focus on numerous areas. First, improved models and data
that enable researchers to control for other variables and estimate the implications for
alternative modes. This would be especially support the findings for the Madrid and
Tokyo cases. Second, the relationship between risk perception and transportation mode
choice can be explored more thoroughly through additional cases such as the February
2004 subway bombing in Moscow, the July- October 1995 metro bombings in Paris, the
New York subway following September 11th 2001, and the November 2008 Mumbai
attacks. It would also be instructive to compare these findings with non- terrorist hazards
and accidents as this would allow for more general risk perception findings to be
revealed, such as the rate at which passengers return to pre- attack mode choices. All such
findings have important economic and policy making implications which have yet to be
explored fully.
18
References
Bernd Rorhmann ( 2000) “ Cross- cultural studies on the perception and evaluation of
hazards.” In Renn, O., & Rohrmann, B. ( eds) Cross- cultural risk perception: A survey of
empirical studies Dordrecht, Kluwer Academic Publishers.
Bontempo, R. N., Bottom, W. P. & Weber, E. U. ( 1997) “ Cross- cultural differences in risk
perception: A model- based approach.” Risk Analysis, 17, 4, 479- 488.
Brenot, J. & Bonnefours, S. ( 1995) research does not support. Approche socio- culturelle
de la perception des risques ( Note SEGR/ LSEES No. 95- 17). IPSN. Referenced in
Rohrmann, B. ( 2000) “ Cross- cultural studies on the perception and evaluation of
hazards.” In Renn, O., & Rohrmann, B. ( eds) Cross- cultural risk perception: A survey of
empirical studies Dordrecht, Kluwer Academic Publishers.
Cha Y- J. ( 2000) “ Risk perception in Korea: a comparison with Japan and the United
States.” Journal of Risk Research, Volume 3, Number 4, 1 October 2000 , pp. 321-
332( 12)
Dake, K. ( 1991) “ Orienting dispositions in the perception of risk.” Journal of Cross-
Cultural Psychology, 22, 61- 82.
Englander, T., Farago, K., Slovic, P., & Fischhoff, B. ( 1986) “ A comparative analysis of
risk perception in Hungary and the United States. Social Behavior, 1, 55- 66.
Fischhoff, B., Slovic, P., Lichtenstein, S., Read, S. & Combs, B. ( 1978) “ How safe is safe
enough? A psychometric study of attitudes towards technological risks and benefits.”
Policy Sciences, 9, 127– 152.
Goszczynska, M., Tyszka, T., & Slovic, P. ( 1991) “ Risk perception in Poland: A
comparison with three other countries.” Journal of Behavioral Decision Making. 4, 179-
193.
Guerrero, P. ( 2002) “ Mass Transit: Challenges in Securing Transit Systems” GAO,
Testimony Before the Subcommittee on Housing and Tranportation, Committee on
Banking, Housing, and Urban Affairs, US Senate.
Hayakawa H., Fischbeck P. S., & Fischhoff B. “ Automobile risk perceptions and
insurance- purchasing decisions in Japan and the United States.” Journal of Risk
Research, Volume 3, Number 1, 1 January 2000 , pp. 51- 67( 17)
Litman, T., ( 2005) “ Terrorism, Transit and Public Safety: Evaluating the Risks” Journal
of Public Transportation, Vol. 8: 4
19
Jacobs, L. & Worthley, R. ( 1999) “ A comparative study of risk appraisal: A new look at
risk assessment in different countries.” Environmental monitoring and assessment, 59, 2,
225- 247.
Jenkins, B. M., ( 2004) “ Terrorism and the Security of Public Surface Transportation.”
RAND Testimony presented to the Senate Committee on Judiciary. RAND Corporation.
Jeong, Y. W., Kwon, H. K., Jeong, T. H., & Kang, K. ( 2002) “ A comparative study about
the traffic accident statistics and the risk perceptions among South Korea, Japan, and the
United States.” In Wang, K. C. P., Xiao, G., Yang, H., Proceedings of the Third
International Conference on Traffic and Transportation Studies. ICTTS.
Kasperson, J. X., Kasperson, R. E., Pidgeon, N., and Slovic, P., ( 2003) “ The social
amplification of risk: Assessing fifteen years of research and theory.” In Pidgeon,
Kasperson, and Slovic ( Eds.), The Social Amplification of Risk. Cambridge University
Press.
Kasperson, R. E., Renn, O., Slovic, P., Brown, H. S., Emel, J., Goble, R., Kasperson, J. X.
and Ratick, S. ( 1988) ‘ The social amplification of risk: A conceptual framework’, Risk
Analysis 8( 2), 177– 87.
Kasperson, R. E. ( 1992), “ The social amplification of risk: progress in developing an
integrative
framework”, in Krimsky, S. and Golding, D. ( Eds), Social Theories of Risk, Praeger,
Westport, CT, pp. 153- 78.
Keown, C. F.: 1989, ‘ Risk perceptions of Hong Kong vs. Americans’, Risk Analysis 9( 3),
401– 5.
Kleinhesselink, R. R. and Rosa, E. A.: 1991, ‘ Cognitive representation on risk
perceptions: Acomparison of Japan and the United States’, Journal of Cross- Cultural
Psychology 22( 1), 11– 28.
Nara, Y. ( 2008) “ A cross- cultural study on attitudes toward risk, safety and security.”
Proceedings of the 12th international conference on knowledge- based intelligence in
information and engineering systems, part II. Springer- Verlag, Berlin.
Plant, J. F., ( 2004) “ Terrorism and the Railroads: Redefining Security in the Wake of
9/ 11” Review of Policy Research, Vol. 21: 3.
Sjöberg, L. ( 1997). Explaining risk perception: An empirical and quantitative evaluation
of cultural theory. Risk Decision and Policy, 2, 113- 130.
Sjöberg, L. ( 1998). Explaining risk perception: An empirical and quantitative evaluation
of cultural theory. In R. Löfstedt & L. J. Frewer ( Eds.), Risk and modern society ( pp.
115- 132). London: Earthscan Publications.
20
Sjoberg, L. ( 1998). Worry and risk perception. Risk Analysis, 18: 1, 85– 93.
Slovic, P. ( 1987) “ Perception of Risk.” Science, 236 ( April), 280- 285.
Sunstein, C. R. ( 2003). Terrorism and probability neglect. Journal of Risk and
Uncertainty, 26, 12– 136.
Waugh, Jr. W. L. ( 2004) “ Securing Mass Transit: A Challenge for Homeland Security”
Review of Policy Research, Vol. 21: 3.
21
Exploring reductions in London Underground passenger
journeys following the July 2005 bombings
Fynnwin Prager,* Garrett Beeler Asay, Bumsoo Lee, and Detlof von Winterfeldt
National Center for Risk and Economic Analysis of Terrorism Events ( CREATE)
University of Southern California
* fprager@ usc. edu
Abstract
We examine the reduction in London Underground passenger journeys in
response to the July 2005 bombings. Using entrance data for London Underground
stations between 2001 and 2007, we incorporate demand and supply factors in a
multivariate time- series regression model to estimate changes in passenger journeys
between different Underground lines. We find that passenger journeys fell by an average
of 8.3 percent for the 4 months following the attacks. This amounts to an overall
reduction of 22.5 million passenger journeys for that period. Passenger journeys returned
to predicted levels during September 2005, yet we find evidence of reduced travel until
June 2006. Our estimates controlled for other factors, including reduced Underground
service provision due to damage from the attacks, economic conditions, and weather, yet
substantial reduction in passenger journeys remained. Our analysis suggests that
heightened risk perceptions are a significant cause of reduced Underground travel,
accounting for around 82 percent of passenger journey reductions following the attacks.
Keywords: Terrorism, Behavioral Responses, Risk Perception, Public Transit,
London Underground.
Introduction
Terrorist attacks target human life and civic infrastructure, and aim to inflict
economic harm through behavioral changes and business interruption. The immediate
effects of terrorism are well documented in the mass media, and the secondary impacts
( changes in behavior) are being evaluated with increasing sophistication. One particular
area of interest for research has been the behavioral responses to terrorist attacks on
transportation systems. Transportation systems are targeted by terrorists because of their
potentially high vulnerability, critical position in the economic system, and most
importantly, large number of individuals.
Past work has developed around potential transportation modal shifts in response
to terrorism events. Looking at the September 11th 2001 attacks, Gordon and colleagues
( 2007) find substantial reductions in air travel well after the initial attacks and estimate a
recovery of the air transit system after 2 years. Ito and Lee ( 2005) find evidence that
shorter distance flights were significantly more impacted than long distance flights,
which lends to the substitution hypothesis, where individuals chose to drive instead of
22
fly, while Beeler Asay and Clemens ( 2009) find evidence that large airports were
impacted proportionately more than small airports - the hypothesis being that individuals
were more inclined to travel to airports with less perceived risk.
Such transport mode shifts can have disturbing consequences. Gigerenzer ( 2006)
studied changes in highway traffic after September 11th, 2001 and estimated that 1,200 to
1,500 additional individuals died in the United States because they substituted flying for
driving ( Blalock & Kadiyali, 2005). This behavioral change appears to be in contrast to
the objective risk of flying versus driving. Sivak and Flanagan ( 2003) estimate the fatality
risk of driving an average- length nonstop flight ( 1,157 km) to be 65 times as risky as
flying.
Other research has examined the impact of terrorist attacks on ground
transportation systems. An unpublished paper by Becker and Rubinstein ( 2004) studies
the changes in bus passenger journeys in Israel following terrorist events. They find that
an attack tends to reduce the number of passenger journeys by about 30 percent in the
first and second months after an attack. Becker and Rubinstein also find evidence of
modal shifts, where individuals choose to ride more taxis after attacks ( 2004). This stands
in contrast to evidence from Spain, where in response to the March 2003 attacks, train
passenger journeys reduced yet no substitution towards car travel was observed ( López-
Rousseau, 2004). Complicating the picture further is unpublished evidence from London
which suggests that following the July 2005 bombings, no significant shift in transport
mode occurred and there was no subsequent increase in transportation accidents or
fatalities ( Fasolo et al, 2010). Moreover, evidence from the 1995 Tokyo sarin gas attacks
suggests that there was no significant reduction in passenger journeys on the attacked
mode ( Prager, Fasolo & Ni, 2010). Such contrasting results indicate the necessity for
robust analysis of each case rather than generalizations.
The causes of behavioral changes following attacks are less clear. The rational
choice model of economic theory suggests that ridership is based on the supply and
demand for each transport mode, with individuals maximizing utility so that aggregate
transportation behavior moves towards an equilibrium point of optimal social welfare.
Economists Becker and Rubinstein ( 2004) argue that risk and fear should be incorporated
into the demand side of this model, especially when considering extreme events such as
terrorist attacks. Hence, individual transportation mode choices are influenced by a range
of risk and reward factors which include the relative prices, rewards, risks, and fears
associated with available transport modes. Two individuals with otherwise identical
preferences could choose different modes if their perceptions of the risk were sufficiently
distinct.
The role of fear in decision making has been explored extensively in the literature
on risk perception ( Slovic, 1987), which argues that individual perspectives on uncertain
future events are often based upon emotionally driven beliefs – sometimes framed in
terms of worry ( Sjoberg, 1998) or dread ( Slovic, 1987; Fischhoff et al, 1978) – as
opposed to calculable risk probabilities. This helps to explain the phenomenon following
September 11th 2001, when US airline passengers appeared to switch to statistically
23
riskier road transit ( Gigerenzer, 2006). The same story appears in studies on tourism and
terrorism, where destinations perceived as riskier are more likely to be avoided
( Ichinosawa, 2006; Fischhoff et al, 2004) and willingness to fly is predicted well by the
level of worry ( Bergstrom & McCaul, 2004).
Changes in risk perception and behavior following terrorist attacks appear to be
neither permanent nor homogenous. Burns and Slovic ( 2007) develop an empirically-derived
dynamic model of behavior, in which individuals are shocked into dramatic
changes before gradually returning to activities at similar levels to those prior to the
attacks. Such dynamism is likely to be exhibited at both the individual and aggregate
levels. Changes in risk perception vary across the population and individuals will avoid
and return to the attacked mode of transport at differing rates.
We examine the case of the London July 2005 bombings. During rush hour on
Thursday, July 7th, 2005, 3 bombings occurred simultaneously on separate London
Underground trains, followed an hour later by a bus bombing. These four bombings
claimed 822 victims, with 52 dead. These attacks, along with the 4 failed attempts two
weeks later, in many ways marked a new era of terrorism on UK soil. They were the first
terrorist strikes in the UK of the post- 9/ 11 era. Both attacks were conducted by
autonomous cells of Islamic extremists that sought to influence UK government foreign
policy by attacking civilians and infrastructure, instilling fear in the general public, and
impacting the economy. The use of suicide bombers and targeting of civilians without
warning contrasted with the incidents surrounding the Northern Ireland conflict, which
until these attacks, was the most recent local terrorism experience for most Londoners.
Moreover, while public transportation systems had been targeted previously, the scale
and intensity of these attacks on the transport system were unprecedented.
We analyze the aggregate London Underground passenger journey data for 2001-
2007, control for supply and demand factors, and still find substantial drops in travel. We
find an overall average 7 percent reduction in passenger journeys for the 4 months
following the incident, a drop of some 22.5 million passenger journeys; however we find
evidence that the reduction could have extended through until June 2006. Our analysis
suggests that the July bombings caused individuals to re- evaluate their transportation
mode choices. Regression results indicate that external factors such as economic cycles
and trends, special events, weather patterns, and transportation prices do not influence the
level of passenger journeys greatly during the period in question. Moreover, the most
plausible influencing factors – service disruption from station closures and other systemic
elements, increased time delays, and lags in passengers returning to the London
Underground following full service resumption - do not explain well the sudden
reductions following both sets of attacks. Therefore, changes to risk perception are a
likely factor in causing model shifts, as suggested by Rubin et al ( 2005; 2007).
Our study also shows that transportation mode choice changes following such
shocks are both dynamic and lasting. These findings raise important questions about the
economic impacts and to what extent these are driven by passenger risk perceptions. The
findings also suggest, however, that supply side factors – such as service reductions,
24
station closures, and time delays – are significantly influential on London Underground
passenger journeys, indicating that policy makers have some level of control over
passenger mode choice and potential economic impacts.
Methods
As stated above, numerous studies have made useful contributions to our
understanding of transportation mode choice responses to terrorist attacks ( Gordon et al,
2007; Ito & Lee, 2005; Beeler Asay & Clemens, 2009; López- Rousseau, 2004; Fasolo et
al, 2010; Prager, Fasolo & Ni, 2010). However, this is not to suggest that each is equally
valid. The sophistication of methods used to estimate reductions in passenger journeys
following the attacks varies substantially. Most use single variable forecasting
techniques, whereby counterfactual results are predicted using only historical data for the
variable in question. These range from comparisons of year- on- year changes for the
given months ( Gigerenzer, 2006; López- Rousseau, 2004; Fasolo et al, 2010) to more
complex Holt- Winters ( Gordon et al, 2007) and ARIMA ( Prager, Fasolo & Ni, 2010)
forecasting models. As with any single variable analysis, there is the danger that omitted
variables may exert influence on the forecasted variable. Though the time- series
approaches are able to capture omitted variables with seasonal trends therein, it is clear
that single variable forecasting has limitations. Another drawback to some these studies is
the aggregation of data to weekly or monthly sets, which neglects the more fine- grained
movements of daily data.
We employ multivariate time- series models to estimate the influence of
exogenous variables on passenger journeys, and in turn predict changes in passenger
numbers resulting from the July attacks. 1 Time series models are generally comprised of
both deterministic and probabilistic elements, the latter being referred to as shocks or
innovations ( Intriligator et al, 1996). In our model, the terrorist attacks are viewed as an
exogenous shock to the London transportation system. Deterministic elements include
pre- existing trends, cycles, and seasonal components, which are controlled for to avoid
inappropriate characterizations of the period. Within this set, both demand side factors,
such as economic factors and other special events, and supply side factors, such as station
closures, are accounted for. A number of models are estimated, which parameterize the
period surrounding the bombings. These models pay attention to the distinct line
groupings being impacted, as well specific time periods to account for special events
occurring. We estimate the impact of the July bombings by comparing the observed
passenger journey numbers with the predicted passenger journey levels. The latter are
obtained using the above model without the post July 7th time variables detailed below.
Regression model and variables
1 Our model is similar Ito and Lee( 2) and Beeler Asay and Clemmens.( 3) While these studies
produce high explanatory power, with adjusted R2 greater than 95 percent, it is possible that further
influential variables are not accounted for, such as the impact of airline bankruptcies on service supply.
Nevertheless, our study incorporates the impact of service supply on passenger journeys.
25
Building upon Ito and Lee ( 2005) and Beeler Asay & Clemmens ( 2009) our
quantitative analysis takes the following form ( variables presented in Table I):
( LU travel) = b 0 + ( demand factors) + ( supply factors)
+ ( time factors) + ( other special events)
+ ( July 2005 attacks factors) +
The dependent variable, LU Travel, represents the number of passengers entering a
London Underground station each day. We observed station data between October 2001
and October 2007. We retrieved from the London Underground Strategic Planning unit in
November 2007. The “ July 7th indicator” variable examines the impact of the July 7th
bombings on the overall dependent variable trend by assigning a “ 0” to all dates prior to
July 7th 2005 and a “ 1” to that date and beyond. To account for the sudden drop in
passenger journeys on the day of the bombings itself, we created an indicator variable of
“ 1” for July 7th 2005 alone, and “ 0” for all other days.
In line with the Burns and Slovic model described above ( 2007), we expected the
perceived risk of ridership to increase immediately following the event and then slowly
fade. As such, passengers would shift away from the attacked transportation mode
immediately following the event, before returning gradually to using the London
Underground system. In this study we do not examine passenger journey volumes on
non- Underground transport modes, and instead treat the group as exogenous to the
model. However, to capture this impact, we created an inverse time trend variable for the
dates post July 7th. Each day from July 7th 2005 onwards was assigned the value 1/ n ( e. g.
July 7th = 1/ 1, July 8th = 1/ 2, July 9th = 1/ 3, etc, with n referring to the number of days
since the attack; Figure 1). We included another indicator variable to capture the impact
of the July 21st 2005 attacks, with a “ 1” assigned to that date and zero to others ( Table II).
We created an indicator variable to capture the impact of the “ Congestion
Charge,” a road- pricing scheme aimed to reduce motor vehicle traffic within central
London. The introduction of the Congestion Charge in February 2003, allied with
substantial investment in public transport, has encouraged many commuters to shift their
mode choice away from private motor vehicles. According to Transport for London, car
traffic decreased by 30 per cent, with overall traffic reducing by 16 percent ( TfL, 2004).
One measurement difficulty in modeling the attacks was the change in congestion
charge on July 4, 2005, which increased the price from the initial 5GBP to 8GBP. Thus it
was not easy identify the difference in the congestion charge effects and the bombing
effects. Nevertheless, we expected that an increase in the congestion charge would
increase the traffic on the Underground, implying our estimated results are less in
magnitude than what they would have been without the increase in congestion charge.
The 2007 “ London Travel Demand Survey” suggests that demand for weekend
travel on the London Underground is lower than weekdays ( 2007). Our data was
26
consistent with that finding, with passenger journeys lower on public holidays. Moreover,
the 2007 survey showed that the purpose for weekend use differs from weekday use.
Work and education trips dominate during weekday peak hours, yet such journeys are
almost non- existent during weekends. Therefore, any weekend and holiday impact
revealed by this indicator largely applied to the most common weekend journey types,
namely “ shopping/ personal business” and “ leisure” as referred to in the Transport for
London ( TfL) report. We interacted the weekend and holiday indicator with the July 7
indicator to reveal the impact of the bombings upon passenger journeys during weekends
and holidays.
A number of economic and trend factors were included in the regression model.
We collected seasonally adjusted data for the monthly Greater London unemployment
rate from the UK Office of National Statistics, which uses the definition recommended by
the International Labor Organization ( ONS, 2005). 2 The population of Greater London
has increased since the turn of the century, increasing demand for London public
transport. To measure population increase we used UK Office of National Statistics
projections for annual mid- year figures ( ONS, 2005). We expected the retail price of
petroleum to positively influence demand for public transportation. We used monthly
retail petrol price data from the UK Department for Business Enterprise and Regulatory
Reform ( 2008). We collected rainfall levels from WeatherOnline, an online
meteorological services company ( 2008). We assume rainfall is more likely to influence
the number of London Underground passenger journeys than other weather factors. 3 The
rainfall data is from Croydon, a suburb 11 miles from central London and the only
meteorological recording station to measure rainfall levels for the majority of days
between 2001 and 2007. Dates with omitted data are estimated through a moving average
of the previous 30 days.
The period since the turn of the century has witnessed institutional changes, with
management of the London Underground network passing from UK central government
to the newly formed Greater London Authority. The average price per journey has
increased during this period, although the differentiation through various measures such
as the Oyster card system means that the distribution of costs is complex. 4 Here, the
average revenue generated per passenger journey is calculated through dividing the total
revenue earned each year by the number of passenger journeys in that year ( TfL, 2008).
An unavailable data point for the year 2001 was extrapolated from those retrieved points
using linear regression.
London Underground stations are periodically closed due to maintenance or other
specific reasons, such as the closures following the July bombings. To account for this, a
2 We did not collect gross regional product indicators are not incorporated due to lack of data availability.
However, we believe unemployment rate measured at the regional level is more instructive than the
national economic indicators. Further, intra- urban passenger travel is more associated with employment
level than the overall production level.
3 Both rainfall and temperature are not included because they are correlated, and rainfall is preferred
because it is more likely to influence London Underground use.
4 The lack of specificity here would be of most concern if the price by individuals were correlated with their
risk perception.
27
variable is calculated that weighs the days that a station is closed – all the days where less
than 100 passing through the gate5 – by the average passenger entrance numbers for that
station between 2001 and 2007. To capture the effects of other service operation, we
include two other variables: one that measures the percentage of full service operation for
each 4- week period, and a second that provides the average excess journey time
passengers faced during each 4- week period. Data for these two variables were from TfL
( 2010). To adjust for seasonal factors, we used 11 month indicators, in which a “ 1” is
assigned to that month, and a “ 0” is assigned to all others. In this set of indicators, we
omit the month of May, as this is a typical month for passenger journeys.
Our regression models had relatively high explanatory power, with an adjusted R2
of around 90 percent for all models. The significance of the majority of the variables
within the model suggests that the remaining noise is caused by insufficiently fine-grained
data or omitted variables. The Durbin- Watson test for auto- correlation was run,
providing a result of 1.71. This suggests that auto- correlation may be apparent, causing
the possibility of underestimated standard- error terms, inflated t- scores, and hence false
positives. To adjust for potential autocorrelation we used Newey- West robust standard
errors. Newey- West robust standard errors assume a heteroskadastic error structure,
which is possibly auto- correlated with some degree of lag. However, a Dickey- Fuller test
on the dependent and independent variables found no evidence of unit- roots. Moreover,
we also conducted an inconclusive Johansen co- integration test. 6
Impact of attacks on London Underground passenger journeys
Following the attacks, London Underground passenger journeys fell sharply
( Figure 2). Figure 2 depicts the change in passenger journeys per week, by the dip to the
right of both vertical lines, which mark the weeks of July 7th and July 21st respectively.
The drop is clearly indicated for all groups of lines, including indirectly affected and
unaffected lines ( Figure 3).
We estimate that weekly passenger journey volumes were reduced from July
through November ( Table III). During this period, there was an estimated average 8.3
percent reduction in passenger journeys when compared with predicted levels, though the
size of this reduction fluctuated. There was an average 14.1 percent reduction for the two
weeks following the July 7 attacks ( July 7 – July 20). The reduction rate then increased to
an average of 18.3 percent for the two weeks following the second attacks ( July 21 –
August 3), before gradually receding to around 6 percent in mid- September. Passenger
journeys moving briefly above predicted levels in September and November. This
amounts to a total mean reduction in passenger journeys of 22.5 million for the 4 months
following the attacks, with a 95 percent confidence range of 14.9– 30.1 million. This is a
conservative estimate because the passenger journey reductions appear to have lasted
5 During a station closure, individuals such as maintenance workers will continue to pass through the
entrance gates. Analysis of the data shows that this figure did not stray above 100 on days where the
passenger numbers were less than 1 percent of average figures.
6 The Johansen co- integration test did not produce sufficient data to compare the trace statistic or the
eigenvalue maximum with the 5 percent critical value. This is possibly due to the presence of
multicollinearity.
28
through until June 2006 ( Figure 2), by which point there was a cumulative mean estimate
reduction of over 38.0 million passenger journeys ( Table III).
Explaining the drop in passenger journeys
The one other potential explanation for the drop in passenger journeys is the
summer school- break period. This pattern is observable in Figure 2 by the dips during the
July and August months of 2003 and 2004. However, Figure 2 also shows that the model
incorporates this summer reduction in the prediction for July 2005, and that the observed
drop following the attacks is more dramatic than the predicted trend. Moreover, the
estimated reductions last through until November, while school age children return early
September and university students return in early October.
Yet as time progresses beyond the date of the attack there are a number of
competing explanations for the reduction in passenger journeys. In this section we
explore these explanations in light of the data, and find that while the supply- side factors
such as station closure contributed to the reduction in passenger journeys, there remains a
significant portion of the reduction attributable to demand- side factors. Our analysis
below suggests that we cannot rule out the hypothesis that the reduction in London
Underground passengers was caused in part by altered risk perception in response to the
July 2005 bombings.
Service disruption due to station closures
The most compelling alternative explanation for the drop in passenger journeys
following the bombings is that station closures for reconstruction caused sufficient
inconvenience for individuals that they switched to different transportation modes, or
simply did not travel. Indeed, the London Underground system was significantly
impacted by the bombings for some time after the event. All stations were closed on July
7th following the attacks. And while the lines not directly affected by the explosions were
reopened the following day, directly affected tube lines were reopened in stages, with full
service returned by August 4 2005.
There are a number of reasons why the drop in passenger journeys cannot be fully
attributed to station closures. First, by breaking the system up into subway lines that were
directly disrupted, indirectly disrupted, and undisrupted by the attacks, we have shown
that all line groupings experienced reductions in passenger journeys. This is apparent in
the Figure 3, as well as the regression model for undisrupted lines in Table IV. However,
the networked nature of the London Underground is also an issue. On the one hand, it
could be argued that closures to one line would encourage passengers to ride substitute
lines, thus offsetting aggregate reductions. For instance, if the specific station could not
be reached directly, other transport modes could substitute the final leg of the journey. On
the other hand, it is plausible that the lack of a complete London Underground journey
could push the individual to choose another transport mode entirely.
29
Either way, our estimates show that the substantial weekly reductions in
passenger journeys remained long after full service was returned to all lines on August 4.
We estimate that passenger journeys were reduced by an average of more than 9.2
percent for each week until early September. This could be explained as a lagged effect
of the initial station closures; individuals who shifted away from the London
Underground may have, for instance, invested in alternative transport modes for a given
period, or may have been unaware of service resumption immediately. However, the
weight of this explanation is diminished further by the fact that reductions appear to have
lasted through until late 2006.
Regression results add further weight to the hypothesis that other factors play a
role in London Underground passenger journey reductions following July 7 2005. The
variables designed to reflect risk perception elements – the “ July 7 indicator” and
“ Inverse days since July 7” – are both significant. This is despite the statistical
significance of the supply- side station closure, service reduction and time delay variables,
which all changed in the expected manner following the July 2005 bombings; the number
of station closures increased, the proportion of total kilometers operated was reduced, and
time delays increased.
We estimate that the contraction of London Underground service following the
attacks caused a 4.1 million reduction in passenger journeys. This accounted for 34
percent of total passenger journey reductions during the first month following the attack.
However, this proportion diminished to 5.5 percent for the second month, and no amount
thereon. This suggests that demand side forces account for around 18.4 million ( 82
percent) of passenger journey reductions in the 4 month period following the attack.
Demand Side Factors
The reasoning in the three previous paragraphs suggests that supply side factors
cannot alone explain the reduction in passenger journeys. Yet it is not clear what demand
side factors can also explain the reduction. The impact of the July 21 bombings provides
some clues to this effect. Despite no additional station closures and no further deaths or
injuries, passenger journeys dropped in the weeks following the July 21 attacks. It
appears that the compounding impact of a second attack, combined with the Police killing
of the innocent Brazilian citizen Jean Charles de Menezes on July 22, caused individuals
to shift away from the London Underground. Despite the importance of such a question
for policy makers, it is impossible to tell from this data which incident had more effect on
transportation mode choice. In any case, it stands to reason that individuals would have
altered their risk perceptions of travel of the London Underground as a result of either of
the incidents.
We do not have the survey data necessary to validate such an explanation, which
would require the same random sample of respondents to be interviewed before and after
the attacks. Nonetheless, three surveys surrounding the July bombings ( Rubin et al, 2005;
30
2007; Goodwin et al, 2005) collectively suggest that the fear of traveling caused British
individuals to travel less. One study suggests changing risk perceptions of terrorist events
was sufficient to cause a small minority to avoid traveling into central London prior to
the July 2005 bombings ( Goodwin et al, 2005). More importantly, some 30 percent of
respondents 11- 13 days after the July 7th bombings declared that they planned to travel
less often as a result of the attacks ( Rubin et al, 2005). In the follow up survey also
conducted by Rubin et al ( 2007), only 19 percent reported traveling less during 2006 in
response to the bombings. These figures are a similar magnitude to our aggregate results.
One possible drawback here is the omission of tourists. However, international
tourists represent only a small proportion of individuals in London at any one time – less
than 5 percent on average – and they are significantly less likely to use public
transportation than London or UK residents ( ONS, 2010). Though these surveys did not
all ask questions regarding travel into central London via the London Underground, the
results suggest that such risk perception explanations cannot be rejected. The use of
aggregate data is another limitation as it restricts deeper exploration of transportation
mode choice following terrorism events. This data does not reveal individual level
decisions, preferences, or risk perceptions.
However, thanks to the different ridership patterns on weekend and weekdays we
are able to assess the influence of trip purpose on post- incident ridership. In another
finding to support the influence of risk perception hypothesis, weekend passenger
journeys took a much longer time to return to predicted levels than the weekday
passenger journeys. This is shown by the significance of the “ weekend and holiday X
July 7th” interaction variable in the regressions results presented in Table V. It stands to
reason that less essential weekend journeys are impacted to a greater magnitude and for
longer if individuals are influenced by the shift in risk perception following the attacks.
This finding highlights the importance of both risks and rewards for transportation mode
choice; where rewards are diminished, risks play a more prominent role. However,
survey data to validate this hypothesis is unavailable. These particular results have
important consequences economically, suggesting that sectors which rely on the non-commuter
travel more prevalent at weekends – such as central London retail and
entertainment industries – were disproportionately impacted ( TfL, 2008).
Beyond the issues of station closures and risk perception, other factors such as
economic cycles, seasonal components or trends are important to consider. As shown in
Table V, most other model variables are significant in the regression results. Yet the data
for some of these variables are monthly, which stands in contrast to the daily data for the
dependent variable and may create noise within the regression estimations. Special events
are also considered. Numerous popular events occurred during the time period observe;
however, we assume that such events are regular enough to be a treated as white noise.
Nonetheless, this could be a source of noise within the model that is not currently
accounted for. The introduction and increased fairs of the Congestion Charge are other
special events within the model, which we would expect to increase the number of
Underground passenger journeys as individuals on the margin shift away from private
road vehicles included in the scheme. Interestingly the Congestion Charge variable
31
carries a negative coefficient in our regression results, which may be the result of the
concomitant improvement in bus service, though may also be the result of collinearity
among similar variables. 7
Conclusions
We estimated the magnitude and length of passenger journey reductions from the
London Underground bombings in July 2005. Both supply and demand- side factors are
then explored in a multivariate time series regression model as explanations for the
reductions. While passenger journey reductions can be attributable to station closures in
part, demand side factors such as economic and trend variables, also appear to contribute
to the reductions. Yet a substantial reduction in passenger journeys remains unexplained.
These findings, when combined with psychological surveys conducted around the event
( Rubin et al, 2005; 2007; Goodwin et al, 2005), suggest that altered risk perceptions
influenced individual transportation mode choice during this period.
The reduction in London Underground passenger journeys following the July
2005 bombings can be explained in part by passengers’ heightened risk perceptions
regarding further attacks. We find an estimated at 22.5 million fewer journeys ( 8.3
percent) for the 4 months following the attacks, though reductions appear to have lasted
into 2006. Our analysis suggests that heightened risk perception is the major demand side
influence on reduce passenger journeys, accounting for around 18.4 million ( 82 percent)
of passenger journey reductions in the 4- month period following the attack.
These findings appear to be similar to that experienced on domestic airlines
following the September 11th 2001 attacks – around 8 percent for the first year and 4
percent for the second ( Gordon et al, 2007). And the reduction is more than the 4- 6
percent observed for the 2 months following the 2003 Madrid attacks ( López- Rousseau,
2004) and the lack of impact following the 1995 Tokyo sarin gas attacks ( Prager, Fasolo
& Ni, 2010). This phenomenon, whereby heightened risk perception following a terrorist
attack leads individuals to shift away from the impacted transport mode, has also been
observed in the aftermath of other recent terrorism events.
This is not to say that these risk perceptions are only based on fear and not on
fact. Clearly, the terrorist event itself is a signal that reasonable people should take into
account when assessing the risks of a future attack. It is likely though, that for a period of
time, the risk perceptions are heightened relative to the actual risks of transportation, thus
leading to a temporary overreaction to the terrorist attack. The fact that the sarin attack in
Tokyo had no or little impact on travel behavior suggests that heightened risk perceptions
are created primarily by very large and dramatic events.
References
Becker GS, Rubinstein, Y. Fear and the Response to Terrorism: An Economic Analysis.
Working Paper, 2004.
7 The VIF test does not show the Congestion Charge variable to exhibit multi- collinearity, though other
variables do fail the test.
32
Beeler Asay GR, Clemens J. Changes in patterns of air travel after 9/ 11. Forthcoming in
Richardson HW, Gordon P, Moore JE ( eds). Global Business and Terrorism.
Cheltenham: Edward Elgar, 2009.
Bergstrom RL, McCaul KD. Perceived risk and worry: The effects of 9/ 11 on willingness
to fly. Journal of Applied Social Psychology, 2004; 34: 1846- 1856.
Blalock G, Kadiyali V, Simon DH. The Impact of 9/ 11 on Road Fatalities: The Other
Lives Lost to Terrorism. Social Science Research Network, Working Paper Series, 2005.
Burns, WJ, Slovic P. The Diffusion of Fear: Modeling Community Response to a
Terrorist Strike. Journal of Defense Modeling and Simulation, 2007; 4: 1- 20.
Department of Business Enterprise and Regulatory Reform. Typical retail and annual
average prices of petroleum products and a crude oil price index, 2008. Available at:
http:// stats. berr. gov. uk/ uksa/ energy/ sa20080424. htm. Accessed on May 15 2008.
Fasolo B, Ni Z, Phillips LD. A study of the impact of the July bombings on Londoners’
travel behavior. Forthcoming, Metropolitan Transportation Center ( METRANS) report
08- 10, 2010.
Fischhoff B, Slovic P, Lichtenstein S, Read S, Combs B. How safe is safe enough? A
psychometric study of attitudes towards technological risks and benefits. Policy Sciences,
1978; 9: 127– 152.
Fischhoff B, de Bruin W, Perrin W, Downs J. Travel Risks in a Time of Terror:
Judgments and Choices. Risk Analysis, 2004; 24: 1301 - 1309
Gigerenzer G. Out of the Frying Pan into the Fire: Behavioral Reactions to Terrorist
Attacks. Risk Analysis, 2006; 26: 347- 351.
Goodwin R, Willson M, Gaines S. Terror threat perception and its consequences in
contemporary Britain. British Journal of Psychology, 2005; 96: 389- 406.
Gordon P, Kim S, Moore, JE, Park J, Richardson HW. The Economic Impacts of a
Terrorist Attack on the U. S. Commercial Aviation System. Risk Analysis 2007, 27: 505-
512.
Ichinosawa J. Reputational disaster in Phuket: the secondary impact of the tsunami on
inbound tourism. Disaster Prevention and Management, 2006; 15: 111- 123
International Passenger Survey. UK Office of National Statistics. Available at
http:// www. statistics. gov. uk/ ssd/ surveys/ international_ passenger_ survey. asp
Intriligator MD, Bodkin RG, Hsiao C. Econometric Models, Techniques, and
33
Applications Upper Saddle River, NJ, Prentice Hall, 1996.
Ito H, Lee D. Assessing the impact of the September 11 terrorist attacks on U. S. airline
demand. Journal of Economics and Business, 2005; 57: 75- 95.
López- Rousseau, A. Avoiding the death risk of avoiding a dread risk: The aftermath of
March 11 in Spain. Psychological Science, 2004; 16: 426- 428.
Office of National Statistics. Monthly Greater London Unemployment Rates. Available
at: http:// www. statistics. gov. uk/ STATBASE/ xsdataset. asp? vlnk= 1389& More= Y.
Accessed on May 15, 2008.
Office of National Statistics. 2004- based revised data, 2005 Available at:
http:// www. statistics. gov. uk/ statbase/ explorer. asp?
CTG= 3& SL= 4127& D= 4127& DCT= 0& DT= 32# 4127. Accessed on May 15, 2008.
Prager F, Fasolo. B, Ni, Z. Analysis of passengers’ reactions to the sarin gas attacks in
Tokyo. Forthcoming, Metropolitan Transportation Center ( METRANS) report 08- 10,
2010.
Sivak M, Flannagan MJ. Flying and driving after the September11 attacks. American
Scientist, 2003; 91: 6– 8.
Sjoberg L. Worry and risk perception. Risk Analysis, 1998; 18: 85– 93.
Slovic P. Perception of Risk. Science, 1987; 236: 280- 285.
Rubin GJ, Brewin CR, Greenberg N, Simpson J, Wessely S. Psychological and
behavioural reactions to the bombings in London on 7 July 2005: cross sectional survey
of a representative sample of Londoners. British Medical Journal, 2005; 331: 606.
Rubin GJ, Brewin CR, Greenberg N, Simpson J, Wessely S. Enduring consequences of
terrorism: 7 month follow- up survey of reactions to the bombings in London on 7 July
2005. British Journal of Psychiatry, 2007; 190: 350- 356.
Transport for London. Annual Reports, 2002- 2007. Available at
http:// www. tfl. gov. uk/ corporate/ about- tfl/ investorrelations/ 1458. aspx. Accessed on May
15, 2008.
Transport for London. Congestion charging: Six months on, 2004. Available at:
http:// www. tfl. gov. uk/ tfl/ downloads/ pdf/ congestion- charging/ cc- 6monthson. pdf.
Accessed on January 10, 2008.
Transport for London, London Underground Performance data. Available at
http:// www. tfl. gov. uk/ corporate/ modesoftransport/ londonunderground/ 1592. aspx
34
Transport for London. Research Briefing; London Travel Demand Survey supplement,
2007. Available at www. tfl. gov. uk/ assets/ downloads/ LTDS- research- supplement. pdf.
Accessed on May 15, 2008.
WeatherOnline Ltd. - Meteorological Services website. Available at
http:// www. weatheronline. co. uk. Accessed October 3, 2008
35
Table I: Variables included in the regression model and expected coefficient signs
Variable Description Expected Coefficient Sign
Passengers in Daily passengers entering London Underground station gates
Time Continuous daily series, Jan 1 2001 to Oct 8 2007 Positive: service improved during period.
July 7th indicator “ 1” for dates July 7th 2005 and after, “ 0” for all others Negative: significant reduction expected
for period following attacks.
July 7th only indicator “ 1” for July 7th 2005, “ 0” for all others Negative: significant reduction expected
on day of attacks.
Inverse days since July
7th
Inverted continuous series beginning July 7th 2005 Negative: significant reduction expected
through this period.
Congestion charge
indicator
“ 1” for dates Feb 17 2003 and after, “ 0” for all others Positive: As road use increases, London
Underground likely to increase.
Weekend and holiday
indicator
“ 1” for weekend and holidays, “ 0” for all others Negative: London Underground less busy
on weekends and holidays.
Weekend and holiday
X July 7th
Interaction between “ Weekend and holiday” and “ July 7th”
indicator variables
Negative: London Underground likely
less busy on weekends and holidays
following attacks.
Unemployment rate Monthly unemployment rate of Greater London area Negative: decreases demand.
Population Annual population of Greater London area Positive: increases demand.
Petrol price Monthly average UK retail petrol price Positive: increases demand.
Revenue per passenger Average price of London Underground passenger journeys Negative: decreases demand.
Proportion of service
operation
London Underground train kilometers operated as a
percentage of full service capacity
Positive: reduces congestion which
increases demand.
Excess journey time Average excess journey time on the London Underground
resulting from delays
Negative: increases travel time, which
reduces demand.
Weighted station
closure
Dates station closed weighted by average passenger
entrances for that station
Negative: increases barriers to entry.
Rainfall Inches of rainfall recorded at Croydon, a suburb of London Positive: reduces substitution modes such
as bus, motor- bicycle and bicycle.
Month indicators Set of indicator variables, one for each month, May excluded Positive in winter, negative in summer.
Table II: Time Variables Equations
= + + + + + +
= 0 if <
+ + if =
+ ( - )- 1 if < <
+ + ( - )- 1 if =
+ ( - )- 1 if >
Where:
= Passengers entering London
Underground stations
= Intercept term
= Demand factors
= Supply factors
= Time factors
= Coefficients on regression variables:
= July 7 only indicator
= July 7 indicator
= July 21 indicator
= inverse days since July 7
= Error term
= Time in days
= July 7
= July 21
Table III: Monthly change in London Underground
passenger journeys ( 2005- 2006)
Four- week period
commencing
Mean
Estimate
Upper
Bound
Lower
Bound
7- Jul- 05 11.2m 9.4m 13.0m
4- Aug- 05 7.3m 5.7m 8.9m
1- Sep- 05 3.9m 2.4m 5.3m
29- Sep- 05 - 0.6m - 2.2m 0.9m
27- Oct- 05 0.8m - 0.5m 2.1m
24- Nov- 05 - 2.5m - 4.2m - 0.8m
22- Dec- 05 7.7m 5.4m 10.1m
20- Jan- 06 1.1m - 0.3m 2.5m
17- Feb- 06 2.7m 1.5m 3.8m
17- Mar- 06 2.2m 1.1m 3.3m
14- Apr- 06 2.8m 1.6m 4.0m
12- May- 06 1.5m 0.3m 2.7m
9- Jun- 06 1.0m - 0.1m 2.2m
7- Jul- 06 0.8m - 0.6m 2.2m
4- Aug- 06 - 0.4m - 1.7m 0.9m
1- Sep- 06 1.8m 0.4m 3.1m
8- Sep- 06 0.8m - 0.8m 2.3m
20- week impact
7- Jul- 05 to 23- Nov- 05 22.5m 14.9m 30.1m
48- week impact
7- Jul- 05 to 8- Jun- 06 38.0m 20.3m 55.8m
38
Table IV: Regressions of daily London Underground
passenger journeys ( 2001- 2007)
Thousands of Passengers All Lines Unaffected Lines Pre July 7 2005
Time 0.18 0.106 1.05***
( 0.78) ( 1.17) ( 3.25)
July 7 2005 indicator - 85.95** - 44.24***
(- 2.05) (- 2.66)
Congestion Charge indicator - 187.30*** - 59.50*** - 69.29
(- 4.71) (- 3.82) (- 1.62)
July 7 2005 only indicator - 112.00 - 128.8
(- 0.28) (- 0.83)
Inverse days since July 7 2005 - 2,041.9*** - 734.5***
(- 4.78) (- 4.50)
July 21 only indicator - 384.9*** - 142.4***
(- 9.50) (- 8.90)
Weekend and holiday indicator - 1,129.4*** - 482.2*** - 1,122.6***
(- 54.62) (- 45.17) (- 49.09)
Weekend, holiday X July 7 2005 - 143.00*** - 34.76***
(- 5.92) (- 3.77)
Monthly unemployment rate - 13.56 - 0.109 47.35
(- 0.68) (- 0.01) ( 1.05)
Annual population 0.002*** 0.00149*** 0.005***
( 2.59) ( 4.13) ( 4.14)
Monthly petrol price - 1.57 0.794 12.93**
(- 0.64) ( 0.82) ( 2.00)
Revenue per passenger - 92.98 - 1062.5 - 9,873.60***
(- 0.05) (- 1.39) (- 3.50)
Weighted station closure - 1.07*** - 1.047*** - 1.13***
(- 9.41) (- 9.04) (- 7.79)
Proportion of service operation 14.42*** 5.864*** - 1.31
( 3.04) ( 3.12) (- 0.25)
Excess journey time 30.32*** 14.18*** - 3.03
( 3.89) ( 4.58) (- 0.32)
Rainfall - 4.28*** - 1.642*** - 5.97***
(- 3.14) (- 3.02) (- 3.43)
January indicator† - 175.00*** - 51.19*** 41.16
(- 4.82) (- 3.64) ( 0.77)
Intercept - 15.68 - 10,810.3*** - 34,986.6***
(- 0.54) (- 4.24) (- 4.20)
N 2,159 2,159 1,335
adj. R- sq 0.887 0.894 0.879
t statistics in parentheses
* p< 0.10, ** p< 0.05, *** p< 0.01, † Other months hidden
39
Table V: Summary statistics for regression analyses
( 2159 observations, indicator variables not presented)
Variable Mean
Standard
Deviation
Minimum Maximum
Passengers (‘ 000s)
All lines 2,425.9 725.2 0 3630.7
Directly affected lines 1,019.9 302.0 0 1499.1
Indirectly affected lines 961.3 301.1 0 1410.4
Unaffected lines 848.3 259.1 0 1316.9
Pre- July 7 2005 2,303.6 681.1 0 3197.7
Unemployment Rate (%) 7.1 0.4 6.2 8.1
Petrol Price ( GB Pence) 82.7 8.1 69.9 97.6
London Population (‘ 000s) 7,432.8 74.1 7,322.4 7,558.4
Revenue Per Passenger ( GB Pounds) 1.4 0.1 1.3 1.6
Rainfall Daily ( cm) 2.1 4.5 0 49.0
Kilometers Operated (% of total) 93.6 3.0 80.2 96.5
Excess Journey Time ( minutes) 7.7 1.6 6.3 16.8
40
Figure 1: London Underground passenger journeys, all lines,
observed and predicted ( 2003- 2006)
41
Figure 2: London Underground passenger journeys, all lines,
observed and prediction 95- percent confidence intervals ( 2003- 2006)
42
Figure 3: Change in London Underground aggregate weekly gate entrances
by line grouping ( 2005)
43
A study of the impact of the July bombings on Londoners’
travel behavior
Barbara Fasolo*, Zhifang Ni and Lawrence D. Phillips
Operational Research Group and Decision Capability Unit
London School of Economics and Political Science
* b. fasolo@ lse. ac. uk
Introduction
On the 7th of July 2005, at the peak of morning rush hour, three bombs
exploded in short intervals on three London Underground trains. Nearly an hour later,
a fourth bomb exploded on a double- deck bus. The bombings killed 52 commuters
and the four suicide bombers, injuring over 7008. This paper presents an analysis of
the impact of these bombings ( 7/ 7) on Londoners’ use of transportation in the
aftermath of 7/ 7 and the risk perception that this use reveals. Analysis of behavioural
reactions to 9/ 11 ( the terrorist attack on US commercial passenger airlines on 11th of
September 2001) suggests that terrorists ‘ strike twice’ – first claiming lives and
damaging infrastructure directly, during the course of the attack, and then indirectly,
through people’s heightened perception of the risk of a repeated attack on the mode
directly attacked, causing a shift to a riskier transport mode ( Gigerenzer, 2006).
However, Spaniards’ reactions to the Madrid train bombings on 11th of March 2004
( M/ 11) did not show evidence of such second indirect damage ( López - Rousseau,
2005). This paper examines whether Londoners’ experience was closer to the US or
Madrid, and finds that although London’s terrorist attack met the conditions for
unleashing similar reactions to M/ 11, Londoners’ experience of 7/ 7 was different
from both US citizens reactions to 9/ 11 and Spaniards’ reactions to M/ 11. We
examine four different explanations for the disparity and offer a policy implication, to
be substantiated by further analysis.
Behavioral reactions to 9/ 11 and M/ 11
The impact of terrorist attacks on travelers’ behavior has been analyzed both
in the aftermath of 9/ 11 ( Gigerenzer, 2004, 2006), and in the aftermath of M/ 11
( López- Rousseau, 2005). These analyses revealed that the attacks had a powerful
effect on travellers. For instance, Gigerenzer ( 2006) found that for a period of one
year after 9/ 11, air travel dropped below the five- year average preceding the event
and was substituted by car travel. Since travelling by car kills more than traveling by
air ( Slivak and Flannagan, 2003), he hypothesized, and found, that such substitution
claimed lives: Highway fatalities increased as a result of drivers avoiding airplanes,
the dread risk ( defined as a low- probability and high damage event).
Gigerenzer’s ‘ dread hypothesis’ rests on three interlinked conditions, and an
8 The terrorists struck twice in London in the same month. The second attack occurred exactly two
weeks later on July 21st: three bombings were attempted on the London Underground, and one on a
bus. None of the main explosive charges detonated, and there were no casualties. It is possible that
both attacks influenced people’s behavior. Due to the short interval between the two attacks, it is
impossible to single out their individual effects. So the subsequent analysis can be viewed as
examining their joint impact, with 7/ 7, the attack that had incurred direct life losses being the leading
factor underlying Londoners’ subsequent behavioral changes.
44
implicit fourth:
1) dread avoidance, evidenced by a decrease in the use of the transportation
mode directly attacked by the terrorists and therefore ‘ dreaded’;
2) substitution, evidenced by an increase in the use of the modes that serve as
the substitute of the mode attacked and dreaded; and 3) increase in fatality.
For 3) to take place, an important implicit condition is that 4) the substitution
mode is riskier, that is, associated with higher fatality rates than the attacked
mode.
This was the case in the US, where after 9/ 11 car travels increased especially on the
rural interstate highways. Interstate highways are the more likely candidates for
substituting within- US air travels; they are also associated with a higher fatality rate
than air travel. Indeed, Gigerenzer found that more people died on the roads following
9/ 11. Immediately following the attack, the number of fatal crashes rose above the
five- year maximum ( 1996- 2000) for each month and remained so for a period of six
months; this number only returned to the five- year average one year after 9/ 11.
Gigerenzer considers this the ‘ indirect’ damage caused by terrorists. Terrorists strike
twice, first physically on people and infrastructure, then psychologically, through
people’s minds.
Interestingly, analysis of Spaniards’ travel reactions to the Madrid terrorist
attack yields different results from the US. Specifically, López- Rousseau found dread
avoidance ( rail usage fell following M/ 11), but no dread- induced substitution ( no
increase in car patronage). Consequently, he found no increase in fatality ( measured
by interannual variations, or the percentage difference between a measure in a given
period and the same period a year earlier, also called year- on- year changes).
López- Rousseau ( also see Gigerenzer, 2006) proposed three explanations for
the apparent disparities between the US and Spain and for the lack of substitution in
particular. First, Spain has a history of terrorist attacks which the US has not. Past
exposure to a risk increases people’s knowledge of the risk, and thereby decreases its
perceived ‘ riskiness’ ( Slovic, 1987). Second, Spain is less of a ‘ car culture’ than the
States. Third, Spain has more developed public transportation systems. These two
suggest that compared to Americans, Spaniards are less likely to replace the affected
public transportation mode ( train travel) as well as less likely to substitute it with car.
On these three accounts, we consider Britain to be more similar to Spain than
to the US, leading us to expect that Londoners’ reactions to 7/ 7 should also show no
evidence for indirect damage in terms of increased fatality, as well as no evidence of
substitution. First of all, the UK has for decades had to deal with terrorist events. For
instance, in 1993, the Provisional Irish Republican Army ( IRA) detonated a truck
bomb in London’s financial district in the City of London, killing one person and
injuring 44. In terms of the efficiency of public transportation systems, London has
well- developed underground and bus networks. The car culture is perhaps most
distinctive in the States. Americans have the highest number of vehicles per capita,
almost twice as many as British or Spaniards9. Besides the attitude, the incentive to
substitute public with private transportation ( car) might even be lower in London than
Madrid, due to the congestion charge introduced in February 2003. This is a daily
9 http:// en. wikipedia. org/ wiki/ Image: World_ vehicles_ per_ capita. svg. Last accessed: 03 July, 2008
45
charge of £ 8 ($ 16) for anyone who drives into the congestion charge zone, which
covers most of central London. A last important aspect that makes 7/ 7 similar to M/ 11
is the fact that both were attacks on ground transit – unlike 9/ 11.
Methodology
We collected five- year transportation data, from 2002 to 2006, from the
transportation authorities of the UK and London, i. e. Department of Transport and
Transport for London. These include: yearly traffic volume of buses10, cars11 and taxis
( as one mode), pedal cycles and powered- 2- wheelers12, weekly traffic volume of
London underground ( in charts), and fine- grained fatality and casualty data by
London borough, by transportation mode, and by month. We analyzed the data by
measuring interannual variations. For fatalities and casualties, we also compared the
data to the average, maximum and minimum of each month of three years before
2005 ( from 2002 to 2004). We measured the ‘ riskiness’ of each transportation mode
by fatality rate in persons killed per million vehicle kilometres, or the number of fatal
injuries divided by traffic volume of each transportation mode. This measurement
allows us to tease out the usage of a mode as a contributing factor of the changes in
the fatality. To examine whether the changes in fatality in 2005 were due to 7/ 7, we
computed 6- month fatality ratios, by using the total fatalities in the second- half of
2005 ( from July to Dec) divided by those in the first half ( from Jan to June), and
again compared this ratio in 2005 to those in the previous three years ( from 2002 to
2004).
Results
The following section presents our results in the logical order suggested by
Gigerenzer’s ‘ dread- hypothesis’: 1) Did avoidance occur? 2) Did substitution occur?
And 3) Did fatalities increase?
1) Did avoidance occur?
The modes of transportation directly affected by the terrorists were the
London underground ( also called the ‘ tube’) and buses. Avoidance would therefore
occur if we found that passenger volumes decreased on both the tube and buses
immediately following the attack of 7/ 7 ( and possibly after the failed attack of 21/ 7),
and gradually returned to the pre- 7/ 7 baseline.
The tube weekly passenger entry data collected from Transport for
London ( Table 1) showed a 12.8% drop in the week immediately following 7/ 7 during
weekdays; the impact on weekends was even larger – a 32% decrease occurred.
Table 1: Interannual variations of London Underground entry: 2005 versus
2004.
10 Buses include buses and coaches.
11 Cars do not include goods vehicles.
12 Powered- 2- wheelers include motor cycles and mopeds.
46
Week commencing
16- Jul 23- Jul 30- Jul 06- Aug 13- Aug 20- Aug
Weekday
entries
- 12.8% - 15.9% - 16.5% - 14.0% - 8.6% - 5.6%
Weekend
entries
- 32.7% - 11.6% - 34.0% - 23.4% - 13.5% - 11.7%
Weekly
total
entries
- 16.5% - 15.1% - 19.7% - 15.7% - 9.5% - 8.4%
Source: Transport for London
As shown in Fig. 1, the decrease probably lasted for at least two months till
mid- September ( the solid line). But since underground patronage had been increasing
robustly since the beginning of 2005 ( the lines were well above the 0% base- line,
which indicates the monthly average of the previous three years), seasonally-corrected
data revealed that the effect might have lasted till early December ( the
dashed line). While these results do not allow us to distinguish between avoidance on
directly hit lines ( which were closed in certain sections until early August) and
avoidance on lines not hit, research that has examined this difference found that
avoidance occurred also on lines not hit ( Prager, Beeler Asay & von Winterfeldt,
2009).
Figure 1: Weekly tube usages in 2005 compared to 2004.
The baseline ( 0%) is the weekly entry of 2004 in the same week. The solid ( red) line shows the actual
weekly entries in 2005 compared to 2004; the dashed ( green) line show the seasonally- corrected
weekly entries. The sudden drop corresponded to the week of 7/ 7. Although the actual weekly entries
suggest that the tube usages recovered in mid- September, the seasonally- corrected data show that the
recovery did not occur till early December. Source: Transport for London.
As for buses, avoidance is less obvious ( Figure 2), mainly because the data
47
currently available is aggregated yearly. The traffic volume of bus and coach in 2005
was comparable to that in 2004. Nevertheless, the year on year % change reveal that
before 2005, bus use had been increasing robustly for two years in a row, but stopped
in 2005 ( 0.33%), and again resumed in 2006 at the 2004 rate. Thus, it is possible that
bus use was affected. To better address this question, we will continue to seek
monthly bus traffic volume data for 2005.
Figure 2: Yearly traffic volume of bus or coach in London.
Million vehicle kilometers 2002 2003 2004 2005 2006
Bus or Coach 534 582 600 602 621
Year on year change (%) 8.99% 3.09% 0.33% 3.16%
The trend lines show that the bus usage in 2005 was comparable to 2004.
Source: Department of Transportation
2) Did substitution occur?
The dread hypothesis posits that travelers avoid the transportation mode
directly hit by the terrorists ( underground and bus) by substituting it with viable
substitutes. Among the possible transportation modes, e. g. pedestrian, pedal cycle,
powered- 2- wheeler, car and taxi ( as one mode), airline, and boat, we considered pedal
cycle, powered- 2- wheeler, and car and taxi as the most likely substitutes for
underground and bus. Table 2 and Fig. 3 show the yearly transportation volume by
transportation mode in London between 2002 and 2006, as well as the interannual
variations of each mode.
The year- on- year changes between 2005 and 2004 ( the green shaded bars)
reveal an increase in the use of pedal cycles and powered- 2- wheelers, but a slight
48
decrease in that of cars and taxis. These data suggest that pedal cycle and two-wheeled
motor vehicles, and in particular the former, probably served as the
substitutes for the tube and buses.
Table 3: Yearly London traffic volume ( in million vehicle kilometres)
and interannual variations ( as %).
2002 2003 2004 2005 2006
Pedal cycles 502 542 523 585 630
7.97% - 3.51% 11.85% 7.69%
2- wheeled motor
vehicles 762 864 809 845 823
13.39% - 6.37% 4.45% - 2.60%
Car & Taxi 26,795 26,376 26,269 26,136 26,398
- 1.56% - 0.41% - 0.51% 1.00%
Source: Department of Transportation
Figure 3: Interannual variations of London traffic volume by mode.
Shaded bars show the change percentages of 2005 compared to 2004 ( shaded bars), which suggest an
increase in pedal cycles and 2- wheeled motor vehicles ( powered- 2- wheelers), but a decrease in cars
and taxis.
3) Did fatalities increase?
49
The last condition of the dread hypothesis requires that fatalities increased as a
result of avoidance and substitution. We examine evidence for this condition by first
comparing the yearly fatalities ( number of deaths) caused by the three modes reputed
to be substitutes to the tube and buses. Note that we also included 2006 data, as this
would allow us to examine whether an increase in 2005 fatality was unique or simply
reflected a general trend towards long term increase.
Table 4: Annual fatalities by transport mode
2002 2003 2004 2005 2006
Pedal Cycle 20 19 8 21 19
Powered 2- wheeler 66 63 47 44 43
Car & Taxi 76 63 54 55 61
Figure 4: Annual fatalities by transport mode.
These trend lines show that the fatality of pedal cycle was the highest in 2005 compared to both the
years before and the year after, a distinctive pattern not shared by the other two modes, i. e. powered- 2-
wheeler and car and taxi.
Fig. 4 shows that the fatality of pedal cycle increased in 2005 compared to
2004, but that of powered- 2- wheeler decreased. This point is perhaps better illustrated
in the interannual variations in fatality ( Fig. 5). It is clear from Fig. 5 that the only
salient increase in fatalities in 2005 happens to pedal cycle. Since as discussed, pedal
cycle is a substitute mode for avoiding the dread of underground and buses, this
increase could provide support for Gigerenzer’s dread hypothesis if we find evidence
that this increase is due to the July bombings. That is, the increase in fatalities should
occur in the second- half of 2005, from July to December, rather than in the first half,
from January to June. To investigate this, we first collected monthly fatality data for
the three transportation modes, plotted below. This is then followed by the half-monthly
data analyses.
Table 5: Interannual variations of fatality in London by mode
2003 v 02 2004 v 03 2005 v 04 2006 v 05
50
Pedal cycles - 5.00% - 57.89% 162.50% - 9.52%
2- wheeled motor
vehicles - 4.55% - 25.40% - 6.38% - 2.27%
Cars & taxis - 17.11% - 14.29% 1.85% 10.91%
Figure 5: Interannual variations of fatality in London by mode
Among the three potential substitute modes of underground and bus, only pedal cycle shows a salient
increase in fatality in 2005 compared to the years before as well as after.
Figure 6: London Monthly fatalities for Pedal Cycles, Powered- two- wheelers and
Cars & Taxis.
51
The solid and dashed lines show respectively the fatality of 2005 and the three- year average between
2002 and 2004. The squares and diamonds are respectively the maximum and the minimum month
fatalities between 2002 and 2004.
The top panel of figure 6 shows that, despite the overall high fatalities in pedal
cycles in 2005 ( the solid line of the top panel) compared to the previous three years
( the dashed line), this increase had already started to take place before the bombings.
The fatalities in April, May and June 2005 were either the same as or higher than the
maximum fatalities for the same month between 2002 and 2004. Therefore, there is
no reason to believe that the increase in fatalities was due to the bombings alone. An
alternative way to capture this is to compute the ‘ 6- month fatality ratio’, or the total
fatalities in the second- half ( between July and December) divided by the total
fatalities in the first- half ( between January and June) of each year. The result is shown
in Table 3.
Table 6: Six- month fatality ratios ( Jul- Dec/ Jan- May) between 2002 and 2005
2002 2003 2004 Average ( 02- 04) 2005
Pedal Cycle 150% 90% 100% 113% 91%
2- wheeled motor vehicles 136% 103% 135% 125% 144%
Car & Taxi 117% 91% 104% 104% 157%
Table 6 shows that the 2005 fatality ratio for pedal cycles is actually smaller
52
( 91%) than the average of the three previous years ( 113%). It follows that the increase
in fatality in 2005 was mainly due to the increase in the first half of the year, prior to
the London bombings. A second insight from this analysis is that while there is no
evidence for an increase in fatalities in 2005 for powered- 2- wheelers and cars and
taxis, this is perhaps because the fatalities decreased significantly a lot in the first- half
of 2005.
Results Summary
Our analyses reveal that following the 7/ 7 bombings, Londoners avoided
underground, and, most likely, buses - the two modes of transportation directly hit by
the terrorists. Londoners thus showed ‘ dread avoidance’, much like American citizens
after 9/ 11 ( Gigerenzer, 2006) and Spaniards following M/ 11 ( López- Rousseau, 2005).
Like Gigerenzer and unlike López - Rousseau, we find evidence for travel mode
substitution, evidenced by the increased use of pedal cycles and powered- 2- wheelers
in 2005 compared to 2004 and 2006. However, unlike Gigerenzer, we find no
evidence that fatalities increased as a result of avoidance and substitution. Thus, our
data fail to support the notion that as a result of avoiding the dread risk, Londoners
suffered a greater loss of life. This is a surprising result, because it shows that
Londoners behaved differently from American as well as Spaniards. In the next
sections, we offer some plausible explanations for this.
Discrepancy between 7/ 7 and M/ 11
First, we turn to the discrepancy between 7/ 7 and M/ 11. This is unexpected,
given that both 7/ 7 and M/ 11 were attacks on ground transportation, and that both
Britain and Spain are comparable on the characteristics proposed by López –
Rousseau ( lack of car culture, efficiency of public transport, history of terrorism). So,
why did substitution occur in London and not in Madrid?
In addition to our findings and those of Gigerenzer ( 2006), avoidance and
substitution were found, as far as we know, in only one other comparable study
( Becker & Rubinstein, 2004). This study found that an attack on a bus in Israel caused
a 30% reduction of bus traffic in the first and second month. At the same time Israelis
used taxis more frequently after the attacks; that is, there was substitution. We
therefore think that the surprising result is the lack of substitution found in Spain,
which we attribute to the different methodologies employed by us vs. López-
Rousseau. First, López - Rousseau analyzed countrywide, rather than city level, data,
as we did. His choice was motivated by the need to compare the results with
Gigerenzer’s, who examined US- wide travel response. We on the other hand focused
on London- wide data – a necessary choice given that the terrorist attacks were
concentrated on London public transport. In our future research, we aim to collect
UK- wide data on traffic and fatalities, to allow for a direct comparison with the
Spain- wide data. Second, López - Rousseau assumed that the substitution mode for
train was car travel. Again, this choice was motivated by the need to compare his
results with Gigerenzer’s, which examined highway traffic. By contrast, we collected
data on all transportation modes, ruling out the unlikely ones ( e. g. boat, airplane),
before focusing on the three most likely substitutes to underground and buses as the
means of transportation within London.
A second crucial factor that distinguishes Londoners’ transportation choice is
53
the fact that Londoners’ travel behavior was heavily influenced by the congestion
charge levied against anyone who drove private vehicles into the congestion charge
zone, which covered most of the central London area ( Zone 1) where the bombings
occurred. This charge was originally introduced in February 2003 at a daily price of
£ 5 and later increased to £ 8 on July 4, 2005, just 3 days before the bombings. This
measure was taken to alleviate congestion within central London. The effect of the
congestion charge on Londoners’ reactions cannot be ignored, and, while current
analysis cannot tease out its direct effect, we have reasons to believe that it has
powerfully shaped how Londoners reacted to the bombings, and in particular their
willingness to substitute means of public transportation.
The congestion charge is likely to have decreased the benefit and increased the
perceived cost of substituting dreaded risk ( underground or bus) with car. As a result,
we expect the substitution from underground and bus to car to be limited, while
substitution to non- chargeable vehicles, e. g. pedal cycles and powered- 2- wheelers to
be more likely. This is what Table 4 shows: the initial introduction of the congestion
charge in 2003 led to a large increase in the use of non- chargeable modes ( i. e., taxis,
buses and coaches, powered two- wheelers, pedal cycles), and decreases in the use of
chargeable modes ( cars, vans, lorries, etc.) This impact was further enhanced, when,
just three days before the bombings, the charge increased from £ 5 to £ 8, producing an
even larger incentive for people to continue using the underground and buses, or to
use non- chargeable vehicles instead.
Table 6: Key year- on- year changes in traffic entering the central London
charging zone during charging hours ( 07.00 – 18.30)
Source: Transport for London
Consideration of the congestion charge allows us to better interpret the
magnitude of the increase in pedal cycles traffic following 7/ 7. This magnitude
( 11.85%, see Fig 3) is even larger than the increase in 2003 ( 7.97%), when the
congestion charge was first introduced. We are therefore confident that pedal cycles
and two- wheeled motor vehicles, and in particular pedal cycles, served as the
substitutes for the tube and buses. In summary, the congestion charge could have
influenced both Londoners’ willingness to substitute and the choice of substitute. It
explains why car was not a substitute, unlike pedal cycle and powered- 2- wheelers.
Discrepancy between 7/ 7 and 9/ 11
54
The second surprising finding pertains to the fact that substitution meant
higher fatalities in the US ( after 9/ 11), but did not mean increased fatalities in
London. We explore the following four explanations for this discrepancy:
1) Could substitute modes used by Londoners have been less risky than the
modes attacked ( i. e. underground and buses)?
2) Could fatalities have increased in some areas but not others?
3) Could casualties, instead of fatalities, have increased?
4) Could fatalities have been prevented by the congestion charge or other
London- specific policy measure?
Explanation 1
One reason why fatalities might not have increased in London could be that
the substitute modes chosen by Londoners are less risky than the modes avoided. To
determine this, we measured the fatality rate of each transportation mode used as a
substitute. This rate is the ratio between the yearly fatalities divided by the yearly
traffic volume of each mode. Table 7 presents the result.
Table 7. Yearly fatality rate in persons killed per million vehicle kilometers
2002 2003 2004 2005 2006
Pedal Cycle 0.0398 0.0351 0.0153 0.0359 0.0302
Powered 2- wheeler 0.0866 0.0729 0.0581 0.0521 0.0522
Car & Taxi13 0.0028 0.0024 0.0021 0.0021 0.0023
Source: Transport for London
In comparison, the yearly fatality rate of the modes directly attacked were
extremely low: 5, 9 and 4 fatalities occurred on the London underground in 2002,
2003 and 200414 whereas the numbers of fatalities for buses and coaches are 7, 5 and
4, respectively. The traffic volumes of buses and coaches are larger than that of pedal
cycles or powered- 2- wheelers ( Fig. 2 and Table 2), and it is reasonable to assume that
Londoners travel more often as well as in longer distances by underground than by
bike. As a result, the fatality rates of the two affected modes are likely to be lower
than pedal cycle and powered- 2- wheelers. That is, the substitution modes chosen by
Londoners are riskier than the modes avoided – just like in the US, suggesting that
this explanation does not hold. Indeed, we find that the fatality rates of all three
transportation modes are lower in 2005 than those in 2002 ( Table 5). The decrease in
powered- 2- wheelers is the largest. That is, the roads are becoming safer to use.
In the most recently published yearly review of the impact of congestion
13 Judged from the fatality rate, these data seem to suggest that cars are safer than buses. This seems to
be a counter intuitive result. The reason is that London Taxi is the safest transportation mode, incurring
only 1 fatality over the four years between 2002 and 2005. We are unable to separate fatality rates for
car and taxi because the traffic volume data are only available for the sum.
14 London underground fatality data are based on financial rather than calendar years, i. e. from 05
April each year to 04 April of the following year.
http:// www. tfl. gov. uk/ assets/ downloads/ safety_ plan_ 2005. pdf, last accessed on 4, July, 2008.
55
charging15, this improvement in road safety was attributed to the London- wide road
safety initiatives over the recent years. In addition to these, Transport for London, the
government body responsible for most aspects of the transport system throughout
London, also introduced interventions including assisting pedestrians and cyclists at
junctions and bus priority measures. These, incidentally, might be another reason why
( 1) road fatality decreased in the period examined, ( 2) car travel failed to increase
after the bombings, ( 3) bus patronage did not fall in 2005 and ( 4) pedal cycles
increased robustly since 2004.
Explanation 2
A second reason why we do not find an increase in fatalities London- wide
could be that we aggregated fatalities across boroughs. Would a different picture
emerge if we collected fatality data by borough and compared the fatalities of
boroughs directly exposed to the bombings and boroughs not directly exposed? We
addressed this by considering the fatalities of the substitute modes ( pedal cycles and
powered- 2- wheelers) for each of the 33 London boroughs separately. Next we
aggregated the data for the three directly hit boroughs ( Camden, City of London and
City of Westminster). Last, we computed the share of the fatalities of these three
directly affected boroughs to the London total. The results are presented below.
2002 2003 2004 Average ( 02- 04) 2005 2006
Pedal Cycle 30.0% 10.5% 25.0% 21.84% 19.0
%
15.8
%
Powered 2- wheeler 1.5% 11.1% 8.5% 7.05% 6.8% 4.7%
Car & Taxi 2.6% 0.0% 0.0% 8.77% 1.8% 3.3%
As shown in Fig. 7, for each of the three transportation modes, the shares of
2005 fatalities of these three directly hit boroughs were always bounded by the one in
2006 and the average of the previous three years, from 2002 to 2004. Hence, there is
no evidence that the fatalities increased in these boroughs in 2005. On the contrary, in
these boroughs the share of fatalities of the two substitute modes, i. e. pedal cycles and
powered 2- wheelers, actually decreased in 2005 compared to 2004.
Fig. 7. % share of the fatalities of the three directly- hit boroughs to the
London total.
15 http:// www. tfl. gov. uk/ assets/ downloads/ fifth- annual- impacts- monitoring- report- 2007- 07- 07. pdf.
Last accessed: 04 July, 2008.
56
Explanation 3
As Gigerenzer and López- Rousseau, we also used fatality data to assess
whether the London bombings imposed a second indirect damage in terms of
substation- induced fatalities. We found no evidence for an increase in fatalities due to
the increased use of pedal cycles and powered- 2- wheelers. A possibility, explored
here, is that substitution led to an increase in road accidents but – perhaps due to the
policy aimed at improving road infrastructure – these accidents did not kill. To test
this we analyzed casualties ( not fatalities) by transportation mode ( pedal cycles,
powered- 2- wheelers, cars and taxi).
As shown in Fig. 8, casualties of powered- 2- wheeler and car and taxi are
below the minimum value of the previous three years ( 2002 to 2004). This is the case
both before and after July 2005. A different and interesting case is offered by pedal
cycle.
Following 7/ 7, there was indeed an increase in pedal cycle casualties in August 2005,
above the minimum of the previous three years. When computing the 6- month
casualty ratio ( i. e. dividing the number of casualties in the second- half of a given year
by the number of casualties in the first half of the same year), we see that this ratio
was 1.09, 1.16, and 1.11 for 2002, 2003, and 2004. In 2005, the ratio was 1.13,
similar to 2006, when it was 1.14. Hence, there is no reason to believe that the
casualties were abnormally high in the second half of 2005.
Fig. 8. Monthly casualties of Pedal Cycles, 2- wheeled motor vehicles and Cars &
57
Taxis in London.
The solid and dashed lines show respectively the casualties of 2005 and those of the three- year average
between 2002 and 2004. The squares and diamonds are the maximum and minimum fatalities in each
month of 2002 and 2004.
Explanation 4
58
Londoners’ substitution of public transport with pedal cycles shows that
Londoners had both a heightened perception of the dread risk ( or else they would
have continued using public transport) and awareness of the costs of substituting
underground and bus with chargeable private transport ( or else they would have
substituted with cars and taxis more, as Becker and Rubinstein).
The absence of substitution- induced fatalities is in our view closely linked
with London roads becoming safer due to Governmental action. While these policy
effects create a challenge in the data analysis of this project, they also offer an
unprecedented opportunity to learn from a ‘ social experiment’. In particular, the
London experience suggests that one way for Governments to mitigate citizens’
reactions to attacks perpetrated by terrorists on public transport is to enhance the
attractiveness of safer transportation substitutes ( or, alternatively increase the relative
cost of riskier modes e. g., charging for car travel) as well as to provide a better public
transportation system which decreases the chance of substitution- induced fatalities.
59
References
Becker Gary, S. and Yona Rubinstein. ( 2004) " Fear and the Response to Terrorism:
An Economic Analysis."( unpublished working paper)
Gigerenzer, G. ( 2004). Dread risk, September 11, and fatal traffic accidents.
Psychological Science, 15( 4), 286- 287.
Gigerenzer, G. ( 2006). Out of the frying pan into the fire: Behavioral reactions to
terrorist attacks. Risk Analysis, 26( 2), 347- 351.
López- Rousseau, A. ( 2005). Avoiding the death risk of avoiding a dread risk.
Psychological Science, 16( 6), 426- 428.
Sivak, M. & Flannagan, M. ( 2003)_. Flying and driving after the September 11
attacks. American Scientist Online ( Jan – Feb).
Slovic, P. ( 1987). Perception of Risk. Science, 236( 4799), 280- 285.
Transport for London. Central London congestion charging impacts motoring, fifth
annual report. Available at: http:// www. tfl. gov. uk/ assets/ downloads/ fifthannual-impacts-
monitoring- report- 2007- 07- 07. pdf. Last accessed: 04 July,
2008.
60
The impact of the 3/ 11 Madrid bombings on consumers travel
behavior
Thomas Baumert*
Chair of the Economics of Terrorism
Universidad Complutense de Madrid
& Universidad Católica de Valencia
“ San Vicente Mártir”
* tbaumert@ ccee. ucm. es
Introduction
Madrid, march, 11 2004. At 7: 39 three rucksack bombs explode in a train
entering Atocha station, Madrid. In quick succession, they are followed by four more
bombs in a train in the Calle Téllez, another on a train that stationed in Santa Eugenia
Station and two more explode in a train near the Pozo del Tío Raimundo. Spain was
duffering the worst terrorist attack in its history.
In the early morning no- one in Spain doubted that it was ETA ( the Basque
terrorist group Euskadi ta Askatasuna) who were behind the massacre, 16 a fact that
was made clear in the rapid succession of institutional and political party statements
condemning the attacks and attributing responsibility to ETA. 17 Only a few experts
detected details which made ETA participation unlikely, but for the moment these
were mere intuitions, which were rejected when the Police told the government that
the explosive used was Titadine, which was that normally used by this terrorist group.
Though it was true that the spokesman for the illegalised Batasuna — the political arm
of ETA— attributed the attack in an early morning radio interview to ‘ agents of
sectors of the Arab resistance‘ ( sic), this hypothesis was rejected by the government,
when CNI ( the Spanish Intelligence Service) intercepted a call from the same
Batasuna spokesman stating that: ‘ We must play for time. Meanwhile, we must blame
the Islamists, later on we’ll see‘. Yet, these statements were not sent out by the press
agencies till 12: 05; some twenty minutes before, the Government had announced the
fact that there were already more than 100 victims.
Nevertheless, the statements by Interior Minister ( Mr Acebes), confirming
‘ without any doubt‘ — and on the basis of information received by State organisations
and Security Bodies— that ETA had been responsible, was backed up almost
immediately by the leader of the Popular Party who indicated that ‘ everything points
16 For a detailled anaysis of the events, see Baumert ( forthcomming), García- Abadillo ( 2004) and
Álvarez de Toledo ( 2004).
17 Although untypical for a research paper, the personal experience of the author might be relevant for
the purposes of this study. The day of the Madrid bombing I went to work as usually, taking both the
metro and bus. That precise morning I a had a meeting with other members of what has later become
the Research Team of the Chair of the Economics of Terrorism of the Complutense University of
Madrid, and of course the main — not to say the only— topic discussed was the attack and its
consequences. At that moment all of us were convinced that it was ETA who had perpetrated the
attack, as the modus operandi was identical to the failed attempt of ETA to blow up a train on New
Year’s Eve. As usually, I went back home taking again the bus and the metro, as I hadn’t taken the car
that morning.
61
to it having been ETA‘ and freshly confirmed by President José María Aznar at
14: 30. However, in this case no specific mention was made of ETA.
Indeed, the Government had a series of strong argumen