=Paper=
{{Paper
|id=Vol-2215/paper4
|storemode=property
|title=Institutional Alarmism and the Damage It Provokes in case of Hydrogeological Disasters: A Simulative Estimation
|pdfUrl=https://ceur-ws.org/Vol-2215/paper_4.pdf
|volume=Vol-2215
|authors=Rino Falcone,Alessandro Sapienza
|dblpUrl=https://dblp.org/rec/conf/woa/FalconeS18
}}
==Institutional Alarmism and the Damage It Provokes in case of Hydrogeological Disasters: A Simulative Estimation==
https://ceur-ws.org/Vol-2215/paper_4.pdf
Institutional alarmism and the damage it provokes in
case of hydrogeological disasters:
a simulative estimation
Rino Falcone and Alessandro Sapienza
Institute of Cognitive Sciences and Technologies,
ISTC – CNR, Rome, Italy
{rino.falcone, alessandro.sapienza}@istc.cnr.it
Abstract— It is common practice for local authorities to So, people tend to create areas with a high concentration of
create weather alerts even when there is no need, in order to inhabitants and structures. When these areas are affected by
protect themselves legally. However, this has a strong negative cataclysms, the damage suffered is enormous: it arises the
effect on the population, involving in a first phase fear and need to identify strategies minimize this problem.
alarmism, and subsequently a drastic decrease of trust in the
In particular, it has been realized how critical the role of the
authority and therefore in what it reports. The catastrophic
result is that in the long-term periods the alert itself loses its authorities is in order to reduce damage, therefore not only in
value, so the population will not respond effectively when it is the interventional phase, but also in the preventive one,
time to do so. leading the population towards the appropriate behavior.
The aim of the authority should therefore be to produce the
The purpose of this work is to provide an idea of the possible most reliable prediction it can, communicating it to the
damage caused by this practice. Therefore, we realized a population so that they make a correct decision.
simulative scenario, in which a population faces a series of events However, even if the quality of weather forecasting has
over time, with the risk of a critical one, while the authority improved over the years, using increasingly effective models,
decides whether to communicate its forecast as it is or to
overestimate it. Trust acts as glue in the close relationship
we are still dealing with forecasts and as such they may be
between authorities and citizens, and then we start analyzing it wrong.
and then showing how its decrease, due to the alarmism, In particular, as Stewart [22] underlines “actions that are based
increases the damage that the population suffers, providing also a on predictions lead to two kinds of errors. One is when an
quantitative evaluation. event that is predicted does not occur, i.e., a false alarm. The
second is when an event occurs but is not predicted, i.e., a
Keywords— trust; social simulation; cognitive agents. surprise. There is an inevitable tradeoff between the two kinds
of errors; steps taken to reduce one will increase the other.”
I. INTRODUCTION This reasoning is now contextualized in the domain of alluvial
The interest in critical hydrogeological phenomena such as disasters. When an event that has not been predicted occurs,
floods has always been high, because of the enormous damage the damage it entails is enormous.
they cause, both in terms of lives and in economic loss. Since it is the duty of the local authorities to inform the
Cunado & Ferreira [4] report that floods represented 40% of population promptly and correctly about what will happen, the
all natural disasters between 1985-2009. Guha-Sapir et al. [11] population will consider the local authority responsible for the
state that in 2013 hydrogeological disasters took the largest damage that occurred, with consequent legal repercussions.
share in natural disaster occurrence (48.2%) and that the most
expensive hydrogeological disaster ever registered happened All this naturally turns the authority away from what is its
in Thailand in 2011, causing US$ 41.4 billion of damages. main task, resulting in the necessity to secure itself. The
This phenomenon is strongly influenced by urbanization: strategy that is implemented is to launch an alert even when
cities act as social hubs, attracting more and more people from there is no real need. This is how the tendency to false alarms
rural areas. Suffice it to say that in 2016 54.5% of the arises, i.e. the choice to overestimate the actual risk. The point
population lived in urban settlements1 with more than 29,000 is that while a false negative involves enormous damage, this
citizens per km2 and these numbers are destined to increase. does not happen with the false positive: if the critical event
that had been foreseen does not occur, there will be no
obvious damage; there are no destructive consequences, nor
1
http://www.un.org/en/development/desa/population/publicati direct repercussions.
ons/pdf/urbanization/the_worlds_cities_in_2016_data_booklet
.pdf
21
However, even this phenomenon has negative effects on long value of the area affected by the event. This is an
periods. If the authority always launches false alerts, in the oversimplification, as cities are often heterogeneous from this
long run the population will no longer trust this and the value point of view, especially if we consider very large areas.
of the alert itself loses its value. In the presence of a true alert, The authors of [25] propose a much more accurate
the population will not respond appropriately and it will suffer approach. They want to realize a simulator able to compute
a very high amount of damage. flood damage on St Maarten Island, one of five island areas of
In this article, we are interested in estimating the quantitative the Netherlands Antilles.
effects of the damage caused by false alerts in the population. Thanks to a GIS software, they estimated the value of each
Through a simulative approach, we will analyze the behavior area as the sum of the building that it contains. They consider
of a population [10] in this context and the long-term effect of many characteristics of the buildings, such as their dimension
false alarms. and the number of floors. Moreover, they classify buildings
We will focus in particular on tangible and direct damages, as according to their use in residential, commercial and
they are more immediately perceivable and economically industrial. Then the authors define 7 damage curves to
quantifiable. Instead, we will not deal with indirect damage, estimate the direct damage to the buildings.
which have an intangible impact and cannot be monetary They also try to estimate tangible indirect damage,
quantified, such as loss of life or psychological trauma. calculated as a fixed percentage of the direct damage, and the
intangible damages, such as anxiety - computed as a function
of flood depth and land use - and loss of productivity –
II. STATE OF THE ART computed as a function of anxiety and income.
The literature has focused on assessing, as accurately as Although these tools are very accurate, they require an
possible, the impact that the weather phenomena have or could excellent knowledge of the territory and anyway the
have on the affected areas. measurements are subject to large variability [15].
The first point to clarify is which part of the damage produced However, all these works limit their focus on estimating the
by an event we want to estimate, thus providing a damage that the event produces. These tools can be very
classification of the various types of damage that are present. helpful, allowing for the individuation of urban solutions that
However, the literature does not converge on a homogeneous can reduce the flood damage. However, although direct
classification. In this paper, we take into consideration the intervention by the authorities is important to prevent damage,
classification proposed by Gentle [8]. Here the damage due to it can have very high costs and take a very long time. On the
natural disasters is divided into 4 types. The main distinction contrary, interventions by individuals are quicker and it seems
occurs between tangible, monetarily quantifiable, and that the citizens' choices can help to reduce the flood damage
intangible damage, which is more difficult to quantify (such as by up to 80%. What we want to do is precisely to link the
loss of life, psychological traumas, etc.). In turn, these are damage suffered by citizens with their choices, which are in
classified into direct, that is the damage caused directly by the turn strongly influenced by authority.
event (damage to roads, buildings, houses...), and indirect, i.e. Our model allows us to study the complex relationship
the secondary damage that the event causes, such as the between the reaction of citizens with what the authority
closure of companies, the decline in tourism, etc. In general, reports, and thanks to this approach we can study the effects of
researchers estimate flood damage mainly focusing on the authority's communications on the damages that occur.
tangible direct damage, since this is the most practical
dimension to estimate economically.
In order to compute flood damage, it is first necessary to III. THE TRUST MODEL
estimate the magnitude of the event and the value of the The trust model used is this work is that of [19], which is an
structures affected. The magnitude is influenced by many adaptation of the cognitive model of trust of Castelfranchi and
variables, however only the most important are taken into Falcone [3]. Trust seems in fact an excellent way to deal with
consideration, such as the flood water level or the duration of information sources [1][2][14][17][24].
the event [9]. This model makes use of the Bayesian theory, one of the most
In general, researchers estimate the damages that an event can used approaches in trust evaluation [18][26], so information is
cause by the means of the simulative approach [20]. For represented as a probability density function (PDF).
instance, in [13] the authors propose a model simulating Each information source S is represented by a trust degree
critical scenarios and evaluating the expected economic loss. called TrustOnSource [5][7], with 0 ≤TrustOnSource ≤1, plus
Here the flood water level is considered as the factor a Bayesian PDF that represents the information reported by S.
indicating the event magnitude. The TrustOnSource parameter is used to smooth the
Olivieri and Santoro [16] express the damage as a product of information referred by S: the more I trust the source, the
a) the average value per unit of a zone, b) the actual extension more I consider the PDF; the less I trust it, the more the PDF
of the territory affected by the disaster and c) the percentage of is flattened. Once an agent gets the contribution from all its
damages suffered. Although they provide a detailed estimation sources, it aggregates the information to produce the global
of the parameters they use in their calculations, they then use evidence (GPDF), estimating the probability that each event is
the average economic value of buildings to determine the going to happen.
22
A. Feedback On Trust 90% of cases nothing happens, so that the citizens who have
Trust is a dynamic value, changing with time depending on the invested have wasted their money.
situation. In this model, starting from a neutral trust level (that After the event, the citizens adjust the trust values of their
does not imply trust or distrust) the agents will try to sources, on the basis of the corresponding performances. We
understand each information source’s reliability repeat this phase 100 times, enough for them to properly
(TrustOnSource), by the means of direct experience for trust evaluate the sources. After that, each citizen possesses a final
evaluations [21][23]. Using the weighted mean, the will capital and it has suffered a given amount of damage. These
perform the feedback on trust. Given the two parameters α and two dimensions are heavily influenced by the authority
β, the new trust value is computed as: strategy on reporting information.
𝑛𝑒𝑤𝑇𝑟𝑢𝑠𝑡𝑂𝑛𝑆𝑜𝑢𝑟𝑐𝑒=α∗𝑇𝑟𝑢𝑠𝑡𝑂𝑛𝑆𝑜𝑢𝑟𝑐𝑒+β∗𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝐸𝑣𝑎𝑙𝑢𝑎𝑡𝑖𝑜𝑛 (1) A. Information sources
α+β=1
In order to take a decision and to maximize the utility of their
TrustOnSource is the previous trust degree and investments, the citizens need to gather information about
performanceEvaluation is the objective evaluation of the what is going to happen[6]. In particular, the citizens can
source performance. This last value is obtained comparing consult two different information sources, reporting some
what the source said with what actually happened. evidence about the incoming meteorological phenomenon:
The values of α and β have an impact on the trust evaluations. 1. The authority, which distributes into the world
With high values of α/β, agents will need more time to get a weather forecast, trying to prepare citizens to what is
precise evaluation, but a low value (below 1) will lead to an going to happen. This is the most competent source,
unstable evaluation, as it would depend too much on the last as it has the means to produce a correct evaluation of
performance. We do not investigate these two parameters in the phenomena, but it is not sure that the authority
this work, using respectively the values 0.9 and 0.1. In order to will faithfully report the forecast.
have good evaluations, we let agents make a lot of experience 2. Citizens’ personal judgment, or self-evaluation,
with their information sources. based on the direct observation and evaluation of the
phenomena. The point is that, usually, the citizens do
not have the means to produce a proper forecast.
IV. THE FRAMEWORK
The simulations were realized using NetLogo [27], an agent- B. Citizens’ description
based framework. A population of citizens, modeled through One of the parameters characterizing the citizens is the trust
cognitive agents and randomly distributed over a wide area, they have in their information sources. This is a dynamic
has to face the risk of a critical event. The citizens have the value, changing because of direct experience (see Section 3.1).
necessity to identify the future weather event on the basis of Each citizen is also characterized by its ability to see and to
their information sources and of the trustworthiness they read the phenomena. We modeled this associating to the
attribute to them. They possess an initial capital to administer, citizens a probability of success, used to produce the forecast
making the correct investments; thus, they need to understand for the meteorological events. In the simulation, we used the
which is the most convenient choice, according to the costs value 50%. Given that there are just two possible choices, it is
and damages related to each decision. The authority informs the equivalent of a random choice.
promptly the citizens about the weather phenomena, providing Further, citizens possess an initial monetary capital; they want
them with its own forecasts. Notice that, being just forecasts, it to save it, but it could decrease in time. Each citizen decides if
is not certain that what it reports is really going to happen. to invest its capital to make security modifications to its own
This depends on the authority’s reliability, its ability to make property, reducing or the possible damage in case of an event.
predictions. However, the authority can decide to overestimate If it does not, it exposes itself to the risk of a possible high
its forecast, raising an alarm when it is not necessary. damage.
The citizens can also evaluate the situation on their own, but
they cannot be as good as the authority in making predictions,
since they do not possess the appropriate means. C. The authority
Then, according to the trust model proposed in Section 3, they The authority’s duty is to inform promptly citizens about what
estimate the probability that each event occurs, considering all is going to happen and to stimulate them to invest in order to
the information they can access and aggregating each single reduce possible damages. We suppose that it is able to inform
contribute according to the corresponding trust value. After all the citizens. As for the citizens, its forecasts are produced
that, they will reason about this information and they will using the probability of success, which may assume the values
decide if to invest or not. 50%, 75% or 100%: the authority is at least as reliable as the
The critical phenomena occur with a 10% probability; when citizens, but it could even produce perfect forecasts.
they do, citizens will suffer 10 units of damage if they did not The point is that, as already said, it is not given that its goal
invest, and 2.5 units of damage if they invested. In the other coincides with its duty. In order to protect itself legally, the
authority could decide to overestimate a forecast, raising an
23
alarm of critical event when it is not necessary. We E. Platform inputs
characterized it with a probability of overestimation, The first thing that can be customized is the number of
determining if it is going to report the truth or not. In the citizens and their probability of success, i.e. their ability in
simulation, it will assume the values 0%, 25%, 50%, 75% and making predictions, and their initial monetary capital. Then,
100%. one can set the value of the two parameters α and β, used for
updating the sources’ trust evaluation.
D. How the citizens decide Concerning the authority, it is possible to change its reliability,
probability of success, and its probability of
Once the citizens gathered information from their sources, the
overestimation. One can also set the critical event’s
processed through trust values and then aggregated it, they are
probability.
able to estimate with what probability there will be a critical
event. Then they need to understand which choice is more
convenient: to invest or not to invest.
V. THE SIMULATION
Each choice has a fixed cost, the investment, and a variable
part, the damage, which depends on the event. The investment The purpose of this simulation is to quantify the damage that
is equal to 1 unit, but they can decide not to invest (0 unit). In the authority’s overestimation effect of events produces in
case of critical event, the damage is equal to 10 if they did not citizens.
invest and to 2.5 if they invested, while it is 0 if there is no Therefore, in the experiment we change the correctness of the
event. authority in making forecasts and its probability to
Table 1 and Table 2 report the cost and damage linked to each overestimate the risk.
decision respectively when there is no event and when there is Each simulation has a fixed duration of 100 events, in which
a critical event. the citizens make experience with their information source and
calibrate the parameters of the model, i.e. the trust that they
Table 1: cost and damage linked to each decision in case of no place in their sources of information.
event At the end of these 100 events, we measure the damage the
To ignore the To take measures citizens suffered and we test their ability to make the correct
problem choice.
Cost 0 1
Damage 0 0
Table 2: cost and damage linked to each decision in case of
critical event
To ignore the To take measures
problem
Cost 0 1
Damage 10 2.5
The citizens compute the probabilistic cost of each choice and
they will make the decision that minimizes the cost:
CostOfInvestment = Investment + (MaxDamage/4)*P(event) (1)
CostOfNotToInvest = 0 + MaxDamage*P(event) (2) Fig. 1. The citizens’ trust on the authority, depending on the authority’s
probability of success and probability of overestimation.
Notice that if we consider the a priori decision, without any The most immediate consequence of alarmism is the
information about what is going to happen, the choice of diminution of trust in authority (Figure 1), at least for this kind
making an investment has a cost equal to 1.25 (Equation 3), of tasks. When the authority does not overestimate its
while the choice of not investing is 1 (Equation 4). forecasts, the trust values are very similar to the authority’s
probability of success. When the probability of overestimation
CostOfInvestment = 1 * 1 + 2.5 * 0.1 = 1.25 (3)
increases, the trust values decrease: the citizens will ignore
CostOfNotToInvest = 0 * 1 + 10 * 0.1 = 1 (4)
what the authority says, since they consider it an unreliable
source
From Equations (3) and (4), we deduce that without
information the best choice is not to invest. The citizens need
to use their information to maximize the utility of their choice.
24
The best performance, i.e. the one that guarantees lower levels
of damage, is obtained when the authority is 100% correct.
Increasing the probability of overestimation, the quantity of
damage increases: it can even reach 2 and a half times the
value of the ideal case.
This huge difference is indicative of the impact of the
authority's communication in preventing damage to the
population.
VI. CONCLUSIONS
The purpose of this article is to provide a quantitative
estimation of the alarmism effects on the population, in case of
Fig. 2. Percentage of citizens’ correct decisions, depending on the authority’s hydrogeological risk.
probability of success and probability of overestimation. Although it is now common practice for local authorities to
overestimate events to protect themselves on a legal aspect, it
is also true that this practice has many negative effects on the
population.
The first effect is that of a decrease of trust in the authority (at
least in this context): since this always reports untrustworthy
information, the population will not trust anymore what it
says, so when there really will be a critical event, the
population will underestimate the alarm.
This therefore leads to the second effect: the decrease in the
performance of citizens. Unable to rely on a reliable source,
their performance inevitably decreases.
The third effect concerns the quantification of the damage. In
fact, agents suffer losses related to their wrong decisions. The
more they are wrong and the higher the damage will be. As we
Fig. 3. Quantification of the damage the citizens suffer in the simulation,
have seen, the damage could even become 2 and a half times
depending on the authority’s probability of success and probability of with respect to the ideal case (100% reliable authority, with
overestimation. 0% probability of overestimation).
In short, although not alarming in case of a critical event may
have immediate catastrophic effects, even the alarmism should
Figure 2 shows the percentage of citizens’ correct decisions, not be underestimated: even if its damage cannot be
depending on the authority’s probability of success and immediately estimate, it can be dangerous for the population
probability of overestimation. The probability of success in the future through secondary effects. This phenomenon
assumes the values 50%, 75% and 100%, represented should be studied more in depth, in order to identifying
respectively in blue, red and green. The probability of solutions that stop it from arising, allowing local authorities to
overestimation assumes the value 0%, 25%, 50%, 75%, 100%, focus on more important goals.
represented in the axis of the abscissas. The results of this study do not want to be exhaustive, but they
As expected, a more skilled authority allows citizens to get a provide quantitative estimates that highlight the critical nature
better performance. The ideal case is when we have a very of the phenomenon and the need for further studies in this
skilled authority (probability of success=100%) that faithfully regard.
reports its forecast (probability of overestimation = 0%).
However it is an impossible case in the real world: even ACKNOWLEDGMENTS
assuming that the authority faithfully reports its prediction,
This work is partially supported by the project CLARA—
every prediction always carries with a degree of uncertainty.
CLoud plAtform and smart underground imaging for natural
Increasing the effect of overestimation, the citizens’
Risk Assessment, funded by the Italian Ministry of Education,
performance decreases to the lower value, which is 50% since
University and Research (MIUR-PON).
the other source (personal judgment) has 50% reliability,
equal to a random choice.
REFERENCES
Figure 3 represents the quantification of the damage the
[1] Amgoud L., Demolombe R., An Argumentation-based Approach for
citizens suffer in the simulation, again depending on the Reasoning about Trust in Information Sources, In Journal of
authority’s probability of success and probability of Argumentation and Computation, 5(2), 2014
overestimation.
25
[2] Barber, K. S., & Kim, J. (2001). Belief revision process based on trust: Systems (AAMAS 2016), Singapore, May 10, 2016, Ceur Workshop
Agents evaluating reputation of information sources. In Trust in Cyber- Proceedings, vol 1578, paper 6.
societies (pp. 73-82). Springer Berlin Heidelberg. [15] Merz, B., Kreibich, H., Thieken, A., & Schmidtke, R. (2004). Estimation
[3] Castelfranchi C., Falcone R., Trust Theory: A Socio-Cognitive and uncertainty of direct monetary flood damage to buildings. Natural
Computational Model, John Wiley and Sons, April 2010. Hazards and Earth System Science, 4(1), 153-163.
[4] Cuñado, J., & Ferreira, S. (2011, July). The macroeconomic impacts of [16] Oliveri, E., & Santoro, M. (2000). Estimation of urban structural flood
natural disasters: new evidence from floods. In Agricultural and Applied damages: the case study of Palermo. Urban Water, 2(3), 223-234.
Economics Association’s 2011 AAEA & NAREA Joint Annual [17] Parsons, S., Sklar, E., Singh, M. P., Levitt, K. N., & Rowe, J. (2013,
Meeting. Pittsburg, PA. Philadelphia, PA: Center for Risk Management March). An Argumentation-Based Approach to Handling Trust in
and Decision Processes, The Wharton School, University of Distributed Decision Making. In AAAI Spring Symposium: Trust and
Pennsylvania Autonomous Systems.
[5] Falcone, R., Sapienza, A., & Castelfranchi, C. (2015). The relevance of [18] Quercia, D., Hailes, S., & Capra, L. (2006). B-trust: Bayesian trust
categories for trusting information sources. ACM Transactions on framework for pervasive computing. In Trust management (pp. 298-
Internet Technology (TOIT), 15(4), 13. 312). Springer Berlin Heidelberg.
[6] Falcone, R., Sapienza, A., & Castelfranchi, C. (2016). Which [19] Sapienza, A., & Falcone, R. A Bayesian Computational Model for Trust
information sources are more trustworthy in a scenario of on Information Sources, in proceedings of the conference WOA 2016,
hydrogeological risks: a computational platform. In Advances in Catania, Ceur workshop proceedings, vol 1664, pp. 50-55.
Practical Applications of Scalable Multi-agent Systems. The PAAMS
[20] Scawthorn, C., Blais, N., Seligson, H., Tate, E., Mifflin, E., Thomas, W.,
Collection (pp. 84-96). Springer, Cham.
... & Jones, C. (2006). HAZUS-MH flood loss estimation methodology.
[7] Falconem, R., Sapienza, A., & Castelfranchi, C. (2015, June). Trusting I: Overview and flood hazard characterization. Natural Hazards Review,
information sources through their categories. In International 7(2), 60-71.
Conference on Practical Applications of Agents and Multi-Agent
[21] Schmidt, S., Steele, R., Dillon, T. S., & Chang, E. (2007). Fuzzy trust
Systems (pp. 80-92). Springer, Cham.
evaluation and credibility development in multi-agent systems. Applied
[8] Gentle, N., Kierce, S., & Nitz, A. (2001). Economic costs of natural Soft Computing, 7(2), 492-505.
disasters in Australia. Australian Journal of Emergency Management,
[22] Stewart, T. R. (2000). Uncertainty, judgment, and error in prediction.
The, 16(2), 38.
Prediction: Science, decision making, and the future of nature, 41, 42.
[9] Grahn, T., & Nyberg, L. (2017). Assessment of pluvial flood exposure
and vulnerability of residential areas. International Journal of Disaster [23] Theodorakopoulos, G., & Baras, J. S. (2006). On trust models and trust
evaluation metrics for ad hoc networks. IEEE Journal on selected areas
Risk Reduction, 21, 367-375.
in Communications, 24(2), 318-328.
[10] Grothmann, T., & Reusswig, F. (2006). People at risk of flooding: why
some residents take precautionary action while others do not. Natural [24] Villata, S., Boella, G., Gabbay, D. M., & Van Der Torre, L. (2011,
June). Arguing about the trustworthiness of the information sources. In
hazards, 38(1-2), 101-120.
European Conference on Symbolic and Quantitative Approaches to
[11] Guha-Sapir, D., Vos, F., Below, R., & Ponserre, S. (2012). Annual Reasoning and Uncertainty (pp. 74-85). Springer Berlin Heidelberg.
disaster statistical review 2011: the numbers and trends. Centre for
[25] Vojinovic, Z., Ediriweera, J. D. W., & Fikri, A. K. (2008, August). An
Research on the Epidemiology of Disasters (CRED).
approach to the modelbased spatial assessment of damages caused by
[12] Latané, B. (1981). The psychology of social impact. American urban floods. In 11th International Conference on Urban Drainage
Psychologist, 36, 343-356. (Vol. 31).
[13] Luino, F., Chiarle, M., Nigrelli, G., Agangi, A., Biddoccu, M., Cirio, C. [26] Wang, Y., & Vassileva, J. (2003, October). Bayesian network-based
G., & Giulietto, W. (2006). A model for estimating flood damage in trust model. In Web Intelligence, 2003. WI 2003. Proceedings.
Italy: preliminary results. WIT Transactions on Ecology and the IEEE/WIC International Conference on (pp. 372-378). IEEE.
Environment, 98.
[27] Wilensky, U. (1999). NetLogo. http://ccl.northwestern.edu/netlogo/.
[14] Melo, Victor S., Alison R. Panisson, and Rafael H. Bordini. "Trust on Center for Connected Learning and Computer-Based Modeling,
Beliefs: Source, Time and Expertise.", in Proceedings of the 18th Northwestern University, Evanston, IL.
International Workshop on Trust in Agent Societies co-located with the
15th International Conference on Autonomous Agents and Multiagent
26