Using Feedback to Conduct Crowd Control
Ivan Taylor
Soni Desai
Centre for Operational
Research and Analysis
NDHQ
Ottawa, Ontario,
CANADA K1A 0K2
Luminita Stemate
Codrin Pasca
Defence Research and
Development Agency
(Valcartier)
Quebec City, Quebec,
CANADA G3J 1X5
(613) 996-0292 (418) 844-4000 ext 4776
ivan.taylor@drdc-rdde.ge.ca luminita.stemate@drdc-rdde.ge.ca
Abstract
Crowd Control is a function generally associated with the police more than the military.
However, the Canadian Forces are occasionally asked to intervene in riot situations
either in Canada, in support to Federal, Provincial and Municipal Governments, or
overseas, during Coalition operations. Thus, there is a need to understand crowd be-
haviour and to determine optimal intervention strategies for crowd control. Arguably the
most famous recent event involving the Canadian Forces was the Oka Crisis in 1990
in which Mohawk Warriors carried out a 76 day standoff and the Quebec Provincial
Police were unable to handle the situation. Another example of situations concerning
governments around the world are unruly protests during visits of foreign dignitaries,
such as the Vancouver APEC Protest, in 1997. Although, the Canadian Forces have
not been employed in these events, they have skills and resources that might be called
upon if a situation gets out of control and must be prepared to deploy on short notice.
A model has been developed that can be used to understand these events in the time
dimension both inside the event and from event to event. The model has been devel-
oped theoretically and “face validated” using data from two case studies. The model is
required to evaluate appropriate tactics such as the employment of non-lethal weapons,
and as a training simulator for strategic and tactical commanders.
1 Introduction
There is a need to take a long-term view of civil disobedience. The apparently suc-
cessful police and military actions to control unruly crowds during one event may have
detrimental effects for future events. Therefore, the Canadian Forces need to have
methods and skills which apply force in the right measure to defuse an event today
while maintaining a balance that will not create new problems tomorrow. One of the
more controversial tools at hand are non-lethal weapons [4, 6]. The effective use of
non-lethal weapons as force multipliers is the subject of much study in the Defence
Research and Development Canada, and in particular in the Centre for Operational
Research and Analysis. As part of this work, the Operational Research Team in Val-
cartier, Quebec has initiated a three-year project that aims at studying crowd dynamics
by means of numerical simulation. Two modelling paradigms have been selected for this
study - System Dynamics and Agent-Based Modelling - with the intention of comparing
their respective advantages and disadvantages when used to simulate crowd dynamics
in confrontation situations. In the first year of this study, the focus has been on data
collection and literature review. As expected, a great deal of recent research is going
on with Agent-Based Modelling applied to crowd behaviour (e.g. {11]). However, there
is a dearth of System Dynamics literature in this area {10}.
The System Dynamics model described in this paper is adapted from a game-theoretic
approach [1] that was originally developed to explain bank runs and currency attacks
[7] using the theory of rational choice [3]. The philosophy behind this model is that
individuals in a crowd make independent decisions to riot if the potential payoff from
rioting exceeds the potential penalties that might be imposed by the controlling forces.
This is implemented in the model through imperfect signaling between the individuals
in the crowd and between the controlling forces and the crowd. These signals are
compared to decision thresholds that employ subjective judgements about potential
payoffs and penalties [5]. One of the features of this approach is that these thresholds
can be solved analytically to obtain a unique equilibrium using Bayes Nash Theory (see
[3]). However, more practically, a feedback technique can be applied using a System
Dynamics model which employs a ‘tit-for-tat’ strategy [2] to obtain a stable equilibrium.
This equilibrium is significant because it demonstrates a balance between the rights of
the individuals for freedom of expression while appropriate force is applied to maintain
control.
Two case studies are described that are of great interest to the Canadian Forces: the
Oka Crisis [12] and the Asian-Pacific Economic Conference (APEC) protest [9]. The
details of these events are provided in Appendix A. The first is based on an aboriginal
standoff in 1990 in which the Quebec Provincial Police were ineffective and the Canadian
Forces were called in to resolve the situation. These standoffs have occurred relatively
frequently over the past twenty years [8] and the Canadian Forces may be involved
in the future. The second is based on a protest against visiting dignitaries during
economic talks in Vancouver in 1997. In this case, the situation was handled by the
Royal Canadian Mounted Police although they came under a great deal of criticism for
their “so-called” indiscriminate use of pepper spray [9]. The Canadian Forces were not
involved in the APEC protest. However, as seen during the World Trade Organization
protest in Seattle in 1999 when the US National Guard had to be called in to quell the
riots, the Canadian Forces need to have contingency plans for handling these types of
protests to ensure that they do not get out of control. We will see that the Oka case
appears to be similar to the common phenomenon of “overshoot and collapse”. The
APEC case appears to show a more “stable oscillation”. However, the possibility of
“expanding oscillation” if the event had gone on longer cannot be ruled out. The novel
aspect of the model that will be developed is that both of these case studies can be
analyzed using the same model. Furthermore, the feedback approach has the ability to
extend the time line to take in more than one event of the same type separated in time
and space.
Section 2 of this paper will provide two behaviour over time graphs that can be envi-
sioned to understand the dynamics of crowd control situations in the “stable oscillation”
case like the APEC Conference and the “overshoot and collapse” case such as the Oka
Crisis. Section 3 will provide a description of the game-theoretic model developed by
Atkeson [1] which will be the framework for the System Dynamics model. The analytical
version of the Atkeson model will be described in Section 3.3 with a numerical example
which demonstrates the common problem with the static approach if the thresholds are
not measured exactly. Section 3.5 provides a System Dynamics version of the Atkeson
framework which includes feedback to solve for the equilibrium numerically. This feed-
back based approach resolves the problems we found with the static approach. In the
concluding section, future work for this three-year project will be discussed such as how
we plan to introduce non-lethal weapons into the model and how this model might be
used as a simulation training game.
2 Behaviour Over Time
It will help us in the hypothesis process to examine the behaviour over time graphs
for the two case studies mentioned. The behaviour will be examined from the crowd's
point of view and plot the cumulative payoffs awarded and penalties inflicted over time
in these two cases. The case studies are described in some detail in Appendix A.
2.1 The Oka Crisis
In Figure 1, we have plotted the cumulative gains of the Mohawk warriors over the
17 major incidents that occurred during the 76 day period of the Oka Crisis in 1990.
It was arbitrarily assumed that an effective tactic by the Mohawk warriors gave them
a payoff of +1 while an effective tactic by either the police or the military gave the
Mohawk warriors a penalty of —1. One can see that the warriors made steady gains
over the first few incidents of the crisis. At the peak, a gun battle between the Mohawk
warriors and the Quebec Provincial Police took place and a policeman was killed. The
police then retreated and the provincial government called in the Canadian Forces. The
Canadian Forces applied overwhelming force and the warriors took continuous losses
until at the end of the time line the warriors surrendered. This appears to be similar
to the classic behaviour mode in System Dynamics of “overshoot and collapse”. This
was not an ideal situation for the warriors or the police although the Canadian Forces
were able to successfully defuse this situation which goes to show the classic axiom of
System Dynamics that growth cannot last forever.
Gain
0 2 4 6 8 10 12 14 16 18
Incidents
Figure 1: Behaviour Over Time Graph for the Oka Crisis
2.2 The APEC Protest
Figure 2 shows the cumulative payoffs/penalties for the 11 significant events analyzed
in Appendix A concerning the APEC protest in Vancouver in 1997. Again, we have
arbitrarily assumed that the payoffs for an effective protester tactic is +1 and the
penalty for an effective police tactic is —1. We can see that there does not appear to be
a clear winner or loser in this case. In fact, this might be close to the situation we wish
to achieve from the point of view of the controlling forces. Also it is not necessarily
negative from the point of view of the protesters because they got some opportunity to
give voice to their dissent.
Gain
°
Incidents
Figure 2: Behaviour Over Time Graph for the APEC Protest
3 Atkeson’s Game-Theoretic Model
3.1 Parameter Definition
Let us define the following parameters of the model.
x will be the signal that is received by an individual in the crowd about the strength
of the police. This will be a normally distributed random variable.
@ will be the strength of the police which will be a Bayesian updated value based
on a strength prior for the police and the signals sent to the crowd.
W will be the positive payoff that is received by individuals that riot if they are
successful. This could be the result of a symbolic gesture or a physical gesture.
L will be the negative penalty inflicted on individuals that riot if they are controlled.
This would most likely be a physical punishment such as the use of a non-lethal
weapon.
a(@) will be a function representing the fraction of the crowd that can be controlled
by the police with strength 6. We will assume that this function is linear with
a(0) = 0 and a(100) = 1.
3.2 The Coordination Problem
When @ is such that a(@) is between 0 and 1, the police cannot control the crowd if
they all riot. If the crowd could perfectly coordinate their behaviour, then their optimal
strategy would be to riot, overwhelm the police, and reap to the payoff. However, we
assume that each individual makes an independent decision about whether to riot or
not based on the signals they receive from the police, and based on their impr
of how the rest of the individuals in the crowd will react to the signals they receive.
ssions
Let’s say for instance that @ = 20. In our linear model, a(@) = 0.2. That is, if more than
20% of the crowd riots, they will overwhelm the police and receive a payoff. If less than
20% of the crowd riots, they will be controlled by the police and receive a penalty. The
rational individual in the crowd will wait to see what the others in the crowd do. The
individual will not take the initiative to act first because he or she cannot guarantee
that enough of the crowd will follow to make the 20% threshold. This is the heart of
the rational crowd’s coordination problem.
3.3. The Threshold Concept for Finding the Equilibrium
The threshold concept that Atkeson proposes works as follows. The crowd applies a
signal threshold x* to the actual signal that they receive from the police about the
police’s strength. This is a trigger such that if an individual sees a signal of « < a*
then he or she will riot. The police have a strength threshold 6* which they use as a
trigger strategy as well. If the strength they wish to apply to the incident is 6 > 6*,
then they will punish the crowd. Thus, Prob(x < 2x*|#*) = the probability of the
individual rioting and Prob(@ > 6*|x*) = the probability of the rioters being controlled
and therefore punished by the police.
To find the equilibrium, we wish to find x* and 6* such that
Prob(x < 2*|6*) = a(0")
and
WProb(@ < &*|x*) + LProb(é > @*|x*) =0
3.4 An Implementation
The simulation shown in Figure 3 and documented in Appendix B utilizes a Bayesian
updating of the strength by employing a normally distributed random variable with
mean
and standard deviation
with x* and 6* solved outside the model using a trial and error approach.
3.5 Results with Atkeson’s Model
A typical behaviour over time graph for this simulation is shown in Figure 4 for the
hypothethical parameters: W = 10, L = 100, mg = 20,59 = 10, s, = 1.
In this case, it is assumed that the payoff is relatively small while the penalty is relatively
ere. Thus, if the police wish to achieve balance, they must be judicious in utilizing
force. In this case, the crowd would be allowed to receive a payoff in approximately
90% of the incidents and the police would impose a penalty in approximately 10% of
the incidents. To implement this, the Atkeson threshold values are 2* = 5.590 and
0* = 7.061.
se
The behaviour over time graph shows the cumulative gain for the rioters for the period
of 100 incidents (to model discrete incidents, we use a ‘work-around’ that there is one
incident every model day). The gain in this case looks fairly well-balanced.
Strength Threshold
Penalty
Signal Deviation
Gain
Gain Rate
ey Updated Strength
Af ys,
Mean Strength
Strength Deviaton
Signal Threshold
Figure 3: Simulation Model for the Atkeson Framework
Gain
400
200
-200
-400
0 10 20 30 40 50 60 70 80 90
Time (Day)
100
Gain : Current
Figure 4: Behaviour Over Time Graph for the Atkeson Model
3.6 Instability in the Long-Term
This short-term result is misleading as we can see from Figure 5 when we extend the
period to 500 incidents. The gain diverges quite considerably from a stable result
because of the well-known problem of imprecise specification of the initial conditions
(i.e because the threshold values were specified to only 4 digit accuracy). This puts into
question the value of the analytical approach to solving the Atkeson model in practice.
Gain
2,000
1,400
800
200
-400
0 50 100 150 200 250 300 350 400 450 500
Time (Day)
Gain : Current
Figure 5: Behaviour Over Time Graph for the Atkeson Model in the Long-Term
3.7 A Feedback Implementation
Figure 6 shows a model and Appendix B provides the documentation of a version of
Atkeson’s framework which employs feedback to dynamically determine the strength
threshold based on any given parameter set. The strength threshold applied by the
police is adjusted downward by a value of if the rioters are successful in an
W-L
event. Recall that if the updated strength (@) is greater than the strength threshold
(@*), the police will use their strength to control the crowd. Thus by adjusting the
threshold downward, the police will be more likely to inflict a penalty on the crowd and
therefore more likely to balance the crowd’s cumulative gain. Conversely, the strength
Wo. if the rioters are controlled.
threshold is adjusted upward by a value of
— P| Strength
djustment Threshold
Gain Signal Deviation Mean Strength
Gain Rate ,
Ee Updated Strength
wh van
Strength Deviaton
Signal Threshold
Figure 6: Simulation Model with Feedback
Figure 7 shows the behaviour over time graph in the long-term for this model. One can
see that there is a sense of stability in this case.
Another useful feature of the feedback approach is that the behaviour stabilizes without
any prior calculation of the strength threshold. That is, after a short ‘burn in’ period
the strength threshold stabilizes at the appropriate level for any given signal threshold.
Figure 8 shows the strength threshold in equilibrium when the model begins with the
strength threshold of zero at the start of the simulation.
Thus, the feedback approach produces a stable solution for the Atkeson model without
complex Bayesian calculations and independent of the initial threshold settings. This is
a very satisfying result created through the use of feedback in the Atkeson framework.
4 Concluding Material
The first year of a three-year study of crowd control modelling using Agent Based
Modelling and System Dynamics has been completed. Although there is a great deal
Gain
200
100
0
-100
-200
0 50 100 150 200 250 300 350 400 450 500
Time (Day)
Gain : Current
Figure 7: Behaviour Over Time Graph for Feedback Model
Strength Threshold
10
75.
5
20.
0
0 50 100 150 200 250 300 350 400 450 500
Time (Day)
Strength Threshold : Current
Figure 8: Solving for the Strength Threshold
of work being done around the world on Agent-Based Modelling, there is very little
relevant work being done in this area using System Dynamics. One promising approach
that was discovered during our literature search was a game-theoretic model which uses
rational choice theory [1]. This threshold-based framework involves a complex analytic
solution technique called Bayes Nash Equilibrium [3]. It was implemented in this study
using a simulation model. However, there were problems with the long-term behaviour
of the framework that may have been caused by the inability to exactly define the
solution thresholds numerically.
As an alternative, a System Dynamics implementation of the framework was developed
that allows for feedback to adjust the strength threshold based on the evolving situation.
This approach has two appealing features. First, the results were stable over the long-
term. And second, the feedback approach adjusts the thresholds dynamically avoiding
the need for the complex Bayesian updating process.
Two case studies were examined from recent Canadian history: the Oka Crisis in 1990
and the APEC Conference protest in 1997. The examination of the behaviour over
time graphs from the point of view of the crowd lent “face validity” to the simulation
model of the threshold framework. It would be highly desirable to conduct more case
studies and attempt to validate the framework by estimating the parameters of the
model based on the data collection. Then apply these parameters in the model to
attempt to replicate the behaviour over time graphs of the case studies.
The examination of the model for various parameter settings to determine the sensitivity
of the cumulative gain to various combinations of input parameters will need to be
conducted. The goal here would be to determine the set of robust tactics that lead to
stable and desirable behaviours.
In the current model, the payoffs and penalties are assumed to be constant values.
Penalties are the primary approach to the use of force in the model. There is a need to
look at various options in applying force and in this case, the penalties would need to be
variable to consider the employment of different types of non-lethal weapons in isolation
and in combination. Similarly, the payoffs should vary depending on the effectiveness of
the crowd’s tactics. It would be beneficial to have a number of independent evaluators
examine the case study materials and rate the payoffs and penalties on a floating scale.
The potential use of this model as the basis of a training simulator will be investigated
in the fullness of time. Tactical commanders would benefit from experimentation with
the model to develop knowledge about how to use feedback to apply levels of force
appropriate to the evolving situation so that the event does not get out of control and
in particular ensuring that their use of force does not make matters worse in the short-
term. Strategic commanders might learn about the systematic longer-term relationship
between events and how the application of force in one event effects future events.
References
[1] Andrew Atkeson. Discussion of Morris and Shin’s “Rethinking Multiple
Equilibria in Macroeconomic Modelling”. Federal Reserve Bank of Minneapolis,
Research Department Staff Report, July 11, 2000.
http://www.atkeson.net /andy /discussion.pdf.
[2] Robert Axelrod. The Evolution of Cooperation. Basic Books, New York, 1984.
[3] Christophe P. Chamley. Rational Herds: Economic Models of Social Learning.
Cambridge University Press, Cambridge, United Kingdom, 2004.
[4] Neil Davison and Nick Lewer. Bradford Non-Lethal Weapons Research Project,
May 2004. http://www.bradford.ac.uk/acad/nlw/.
[5] Mark Granovetter. Threshold Models of Collective Behaviour. The American
Journal of Sociology, 83(6):1429-1443, 1978.
[6] Terry Kiss. Non-lethal weapons, July 2005.
http://www.au.af.mil/au/aul/bibs/soft /nonlethal.htm.
[7] Stephen Morris and Hyun Song Shin. Rethinking Multiple Equilibria in
Macroeconomic Modelling. NBER Macroeconomics Annual 2000, May, 2000.
http: //hyunsongshin.org/www/nber.pdf.
[8] Canadian Press. Timeline: Aboriginal standoffs in Canada, April 2006.
http://www.ctv.ca/servlet /ArticleNews/story /CTVNews/20060421/
aboriginal_standoffs_060421/20060421/.
[9] W. Wesley Pue. Pepper in our Eyes: The APEC Affair. UBC Press, Vancouver,
2000.
[10] Ivan Taylor. Models of crowd control a query on the SD mailing list, May 2005.
http://www.ventanasystems.co.uk/forum/viewthread.php?tid=2349#pid6398.
[11] IMA “Hot Topics” Workshop. Agent based modelling and simulation, November
2003. http://www.ima.umn.edu/complex/abstracts/11-3abs.html.
[12] Geoffrey York and Loreen Pindera. People of the Pines: The Warriors and the
Legacy of Oka. Little, Brown and Company, Toronto, 1991.
A Case Studies
Table 1: The Time Line for the Oka Crisis
There are more than 100 officers and Mohawks fear more are
to come so they build a wall of tires and set them on fire to
prevent other SQ officers from arriving. The SQ officers begin
to move forward using tear gas and concussion grenades. A
gun battle starts. Hundreds of rounds are shot but the battle
lasts no more than a few seconds. SQ Marcel Lemay is killed.
Raphals recounts, The SQ officers launched an ill-planned attack
on the Mohawk barricade at Kanesatake. Their own tear gas
blew back in their faces, one police officer was shot and killed-
perhaps by his own side- and they were forced to retreat. In the
aftermath the Mohawks fortified their barricade with crushed
SQ cruisers, topping it off with a sign reading, “They came,
they saw, they ran”. The SQ had retreated, leaving most of
their vehicles near the barricade. The Mohawks destroyed police
cars and took control of a bulldozer to reinforce their barricade.
The SQ watched the Mohawks for the rest of the day from a
helicopter. Mohawk Warriors at the Kahnawake reserve 29 km
southwest of Montreal set up a blockade at the Mercier bridge in
the morning in support of the Kanesatake Mohawks. The bridge
is used by up to 60,000 commuters every day from South Shore
communities to Montreal.
Date and Description of Main Incident Payoff
11 Mar 90: Kanesatake Mohawks set up a blockade in an effort | Blockade (ef
to stop the town of Oka from expanding a golf course on land | fective for
the Mohawks claimed contained a cemetery. Mohawks).
11 Jul 90: The Sortie du Quebec (SQ) police are called in. Just | Tear gas (not
before dawn the SQ arrived in rental cars and vans. The SQ fires | effective for
tear gas after asking to speak to a Mohawk spokesperson and | SQ). Gun battle
are not satisfied when a woman says she is the spokesperson. | (effective for
Mohawks). De-
stroying police
cars and seizing
a bulldozer (ef-
fective for Mo-
hawks). Mercier
Bridge — blocked
(effective for
Mohawks).
Table 1: The Time Line for the Oka Crisis (Continued)
Date and Description of Main Incident
July 90: The Mohawk Warriors join the Kanesatake Mohawks
at the barricades. The Quebec government orders the SQ to
erect its own barricades on the roads leading to the municipality
of Oka and the Kanesatake reserve. The police set up roadblocks
5 km outside of Oka. Both sides try intimidation tactics during
the night. They send men into the woods. Sometimes officers
or Mohawks go in silence to spy and other times they create
noise to let the other side know they are watching. The police
have greater resources and always have fresh crews to watch the
barricade. The Warriors have limited resources and begin to
suffer sleep deprivation.
Payoff
SQ _ barricades
(effective for
SQ). Night
intimidation
(effective for
SQ).
16 Jul 90: SQ stop food shipments at the roadblocks they have
set up. The circulation of food and medical supplies to the com-
munity of Kanesatake experience some delays. Some Mohawks,
are refused access, and leave the food they had brought for their
family at the barricades. The Mohawks ask for a team of ob-
servers from human rights groups across Canada to be present at
the barricades, to prevent the police from violating their rights.
The Red Cross goes behind the barricades and identifies 200 peo-
ple needing special attention. Complaints against the police are
lodged with the Human Rights Commission accusing the police
of harassment, making racial slurs and gestures, and detaining
people. These incidents make natives afraid to cross the barri-
cades to buy food or seek medical supplies.
Blocked food
shipments (ef
fective for SQ).
26 Jul 90: The police report that the bullet that killed Cpl.
Lemay on 11 Jul 90 did not come from a police weapon.
7 Aug 90: Many Mohawks leave Kanesatake.
Mohawks leave
(effective for
SQ).
14 Aug 90: Chief of Defence Staff announces the deployment of | Army — coming
units to Kanesatake and Kahnawake relieving SQ. The Canadian | (effective for
Forces announce that as many as 4,400 soldiers, backed by ar- | SQ).
mored personnel carriers and heavy weapons would be deployed.
Table 1: The Time Line for the Oka Crisis (Continued)
Date and Description of Main Incident
Payoff
20 Aug 90: The arrival of the army. 33 troops arrive around
0800. The army moves one of the SQ barricades 400 meters
closer to the Mohawks to improve the barricades strategic value
to the Army. The Mohawks apparently misunderstand this as
an act of aggression. The negotiations are fragile and every time
the Army pushes the barricades closer, the talks break off. When
the SQ was in Oka the barricades were sometimes separated by
as much as 1.5 km. Now at some points the barricades are as
close as 5 meters from each other.
Barricades move
closer (effective
for Army).
24 Aug 90: Four smoke bombs are set off by the Canadian
Forces in the morning.
1 Sep 90: At 1300, some soldiers climbed over the north, east,
and west perimeter and made their way into the pines. This was
a very tense time. Warriors were trying to provoke the soldiers
into firing the first shot. It was only strict discipline that pre-
vented this situation from ending in a bloodbath. The Warriors
would not retreat. They were ready to stand their ground. It was
finally decided that they would retreat into the treatment center.
The night remained tense. The Warriors took turns guarding the
entrance of the treatment center. The area was illuminated with
high-powered searchlights from across the road. A cacophony of
eerie war whoops amplified through Mohawk loud speakers, heli-
copters hovered above, flares going off only increased the tension
on both sides.
Warriors retreat
into Treatment
Center (effective
2 Sep 90: Barricades are brought down at Kanesatake.
for Army).
Barricades
brought down
(effective for
Army).
6 Sep 90: Mercier Bridge is re-opened by the Army after 62
days of Mohawk occupation. The Army completely surrounds
the Mohawks at the Treatment Center in Kanesatake using razor
wire.
Mercier Bridge
re-opened — (ef-
fective for
Army). Razor
Wire surrounds
Mohawks in
Treatment Cen-
ter (effective for
Army).
Table 1: The Time Line for the Oka Crisis (Continued)
Date and Description of Main Incident
Payoff
8 Sep 90: An incident occurred when four soldiers crossed the
razor wire scouting a Mohawk in the forest, when they stumbled
upon a warrior asleep in a trench. A battle erupted and two
soldiers are wounded. The warrior is overwhelmed and beaten.
Beating of Mo-
hawk (effective
for Army).
13 Sep 90: The phone lines are cut. The only line the Warriors
had was the Hot Line to the negotiation office of the Army.
Light towers are set up to see the Mohawk side of the razor wire
at night. Warriors put up blankets in the woods so the Army
cannot see what they are doing. Soldiers shine powerful lights
into the detox center from 110-foot towers. The army steadily
pushes its side of the barriers forward, tightening their circle of
control; they jam the cellular phones of journalists inside. The
Warriors direct loudspeakers at the troops, the music alternating
between traditional native songs and the Rolling Stones:“Wild
horses cant drag me away ...”. Later the Warriors find a place
where cellular phones are not effected by the blackout and call
it the “Phone Booth”.
Phone lines cut
and light towers
used (not effec-
tive for Army).
Sep 90: One night a Warrior reached over the razor wire with a
long hook and yanked on the Armys tripwires, setting off flares.
The soldiers thought a mass breakout was starting. The Army
was not about to sce this situation get any worse. An officer
then gave an order to “lock and load” and shortly after a shot
was fired in the air.
Set off — trip-
wires (effective
for Mohawks).
Army locks and
loads (effective
for Army).
26 Sep 90: At the local detox center in Kanesatake, the twenty
Warriors and their families who had been making a last stand,
lay down their arms and turn themselves in. The last barricades
are taken down. The Mohawk Warriors burned two fires when
they surrendered.
Mohawks — sur-
render (effective
for Mohawks
and Army).
Table 2: The Time Line for the APEC Protest
Date and Description of Main Incident
Payoff
17 Nov 97: Students begin ‘laying siege’ for the summit by
building a tent city outside the student union building. Named
‘Democracy Village’, it becomes part of the security zone for the
summit. The eleven students, who set up a tent city, wrote slo-
gans on windows and rearranged rocks to spell out a message.
The students are warned that if they do not leave they will be ar-
rested. Seven students leave and the remaining four are charged
and arrested.
Arrests (effec-
tive for RCMP).
Table 2: The Time Line for the APEC Protest (Contin-
ued)
Date and Description of Main Incident Payoff
21 Nov 97: The graduate student’s union at UBC raises the | Flags raised
Tibetan national flag, a symbolic image of independence, atop its | (effective for
student centre and hangs two anti-APEC banners. The RCMP | protesters).
removes the flag on the morning of 25 Nov 97.
24 Nov 97: A protester, Law student Craig Jones, places pa- | Signs raised
per signs reading “Free Speech”, “Democracy”, and “Human | (effective for
Rights” on fences surrounding Green College. At midnight, po- | protesters).
lice remove the signs.
25 Nov 97: At 0750, Craig Jones displays signs that state “Free
Speech”, “Democracy”, and “Human Rights” on the sidewalk on
two coat racks in view of the motorcade route that will be trans-
porting the APEC delegates to UBC. An RCMP officer immedi-
ately orders the protester to move the signs off the sidewalk and
onto the grass 12 feet behind the security fence. The protester
complies. After approximately 10 minutes, the same officer in-
forms the protester that the signs cannot remain but he can. The
protester does not remove the signs and the signs are grabbed
from him. He is handcuffed and arrested. At approximately
0830, Mike Thoms briefly displays a textile banner. Police tell
him he cannot do this and the police seize the banner. Police tell
other individuals wishing to display signs on the Green College
side of the security fences that signs are not allowed.
Signs and ban-
ner removed
(effective for
RCMP). Craig
Jones arrested
(effective for
RCMP).
25 Nov 97: A massive rally peaked on campus at 1230 in a
melee of pepper spray and arrests. It is alleged that individ-
uals carrying cellular telephones or amplifying equipment were
arrested and that women, but not men were strip searched by
RCMP officers. At 1240, protesters reached the 12-foot fence
separating them from the wide secure zone around the museum
and several protesters grabbed the chain link and tore a section
down. The falling fence caught a group of activists and media
underneath before waiting police moved in with pepper spray
and police bikes, beating back the protesters. Shortly after the
security fence is repaired, a group of about 40 activists risk ar-
rest by forming ranks of four and walking slowly into the police
line. The police douse the first rank with pepper spray and they
spray the substance over protesters heads and into the crowd.
Police then threw two-dozen empty bottles back over the fence.
The confrontation ended after about a one-hour standoff.
Rally (effective
for Protesters).
Pepper spray
and arrests
(effective for
RCMP). Bring-
ing down part of
chain link fence
(effective for
Protesters).
Table 2: The Time Line for the APEC Protest (Contin-
ued)
Date and Description of Main Incident Payoff
25 Nov 97: Activists moved on to a trio of roadblocks where | Roadblocks
they sat on the three routes leading away from the UBC museum. | (effective for
With little warning, the police break up one of the roadblocks | Protesters).
by pepper spraying the 50 activists, as well as onlookers and the | Pepper spray
media, forcing the crowd about 100 meters from site of the mo- | (effective for
torcade route. Along the same route the APEC motorcade had | RCMP). Arrests
taken in the morning, some 30 protesters sat on the pavement. | (effective for
Another 200 protesters and onlookers stood on the median in | RCMP).
the road, offering support. By day’s end, 49 people are arrested.
Police also search and release four men with Indonesian accred-
itation who were photographing protesters.
B_ Model Documentation
B.1 Atkeson Model
(01) FINAL TIME = 500, Units: Day, The final time for the simulation.
(02) Gain = INTEG (Gain Rate,0), Units: **undefined**
(03) Gain Rate = IF THEN ELSE(Updated Strength<Strength Threshold, Payoff,
Penalty ), Units: **undefined**
(04) INITIAL TIME = 0, Units: Day, The initial time for the simulation.
(05) Mean Strength = 20, Units: **undefined**
(06) NOISE SEED = 23445, Units: **undefined**
(07) Payoff = 10, Units: **undefined**
(08) Penalty = -100, Units: **undefined**
(09) SAVEPER = TIME STEP, Units: Day, The frequency with which output is
stored.
(10) Signal Deviation = 1, Units: **undefined**
(11) Signal Threshold = 5.590, Units: **undefined**
(12) Strength Deviaton = 10, Units: **undefined**
(13) Strength Threshold = 7.061, Units: **undefined**
(14) TIME STEP = 1, Units: Day, The time step for the simulation.
(15) Updated Strength = RANDOM NORMAL(-100, 100 , (Mean Strength/Strength
Deviaton’2+Signal Threshold /Signal Deviation2)/((1/Strength Deviaton
‘2)+(1/Signal Deviation’2)) , SQRT(1/ (1/Strength Deviaton*2+1/Signal
Deviation’2)), 0), Units: **undefined**
B.2 Feedback Model
(01) Adjustment = IF THEN ELSE(Gain Rate<0,
-Payoft/(Payoff-Penalty) -Penalty /(Payoff-Penalty)), Units: **undefined**
(02) FINAL TIME = 500, Units: Day, The final time for the simulation.
(03) Gain = INTEG (Gain Rate,0), Units: **undefined**
(04) Gain Rate = IF THEN ELSE(Updated Strength<Strength Threshold, Payoff,
Penalty ), Units: **undefined**
(05) INITIAL TIME = 0, Units: Day, The initial time for the simulation.
(06) Mean Strength = 20, Units: **undefined**
(07) NOISE SEED = 23445, Units: **undefined**
(08) Payoff = 10, Units: **undefined**
(09) Penalty = -100, Units: **undefined**
(10) SAVEPER = TIME STEP, Units: Day, The frequency with which output is
stored.
(11) Signal Deviation = 1, Units: **undefined**
(12) Signal Threshold = 5.590, Units: **undefined**
(13) Strength Deviaton = 10, Units: **undefined**
(14) Strength Threshold = INTEG (Adjustment, 7.061), Units: **undefined**
(15) TIME STEP = 1, Units: Day, The time step for the simulation.
(16) Updated Strength = RANDOM NORMAL(-100, 100 , (Mean Strength/Strength
Deviaton’2+Signal Threshold/Signal Deviation’2)/((1/Strength Deviaton
‘2)+(1/Signal Deviation’2)) , SQRT(1/ (1/Strength Deviaton*2+1/Signal
Deviation’2)), 0), Units: **undefined**
