Modeling the 2005 Hatlestad Slide
Ying Qian
Shanghai University, School of Management
Leire Labaka Zubieta
University of Navarra, Industrial Management Department
Peter Lango
University of Bergen, Institute for administration and organization
Jose J. Gonzalez
Agder University, Centre for Integrated Emergency Management
ABSTRACT
It has long been ized that the of ies requires that response organizations act flexibly,
an “emergent organization” to better manage the fact that disasters do not follow scripts. Nevertheless,
recent research shows that crisis response organizations prefer to follow patterns adequate for normal situations.
Arguably, the resi to become an ion could be related to poor understanding of how to
move from di: ization to self- ization. We present a system dynamics model describing the transition
from di: ization to self- ization in the of the fatal Hatlestad landslide in Norway. We
suggest that the causal structure of the system dynamics model describing the Hatlestad case should be
considered a candidate for an emergent “middle-range theory” describing the management of disorganization in
emergencies. We propose specific data collection to test the candidate theory.
Keywords
. of i t izati theory building, system
dynamics
INTRODUCTION
Disasters can be difficult to predict and often come by surprise. Emergency organizations can hardly prepare for
all types of disasters. Dynes and Quarantelly (1976) pointed out from a compilation of 36 studies published
1957-1973 that organizations trying to hold on to traditional plans and dures run into problems when an
emergency occurs and do a worse job at responding than those who move to greater flexibility or resiliency.
Despite this early and well-known insight, a recent study based on questionnaire responses from an international
sample of emergency managers shows that “response organizations prefer to follow the general protocols of
communication set-up for normal situations also in the event of a crisis” (Palttala, Boano et al. 2012, p. 5).
Crisis management is extremely complex, even becoming increasingly more complex (Boin and Hart 2008).
Several theory-based approaches haven used and new approaches are emerging. Gilpin and Murphy (2008, p. 4)
argue that the contingency, uncertainty, and happenstance; the unexpected confluence of unrelated events; and
the destabilizing influence of rapidly changing circumstances are central factors in complex crises. Hence, an
approach is needed that “maintains a vision of the changeable and complex nature of crises, and it looks for ways
to operate within that very real-world environment of confusion, unforeseen events, and missing information.”
Accordingly, complexity theory, “the study of interaction processes within complex systems, including social
systems such as organizations” (Gilpin and Murphy 2008, p. 6) ought to play a central role to advance our
ding of crisis Gilpin and Murphy are not the only proponents of an approach based on
complexity theory. Other recent research implies that crisis operations need to be “multi-organizational, trans-
jurisdictional, polycentric response networks.” They demand lateral collaboration and coordination, not top-
down command and control (Boin and Hart 2008, p. 147).
Simpson and Hancock (2009) point out that most past work on emergency response has addressed well-
structured problems of emergency services. Such studies assumed implicitly that an “Established” organization
(Dynes and Quarantelli 1976) responds based on detailed planning. However, disasters tend to disrupt plans and
impair the very same organization in charge of the emergency response. Simpson and Hancock analyze future
research opportunities and conclude that “a single thread unites the particular areas of research opportunities
discussed earlier in the paper: each is a problem of working in a disorganized environment” (Simpson and
Hancock 2009, p. $136). They go as far as characterizing management of disorganization as an important
operational research growth area for the next 50 years (Simpson and Hancock 2009, p. $126).
The case for a new approach based on complexity theory is strong. However, complexity theory is a vision
involving many and intricate aspects related to interaction, holism, emergence and chaos (Byme 1998). Related
concepts are self-organization, complex adaptive systems, and co-evolution. Complexity theory attempts a
generic, overall explanation of a whole range of phenomena - it is a general theory (in the sense of Schwaninger
and Grésser, Schwaninger and Grosser 2008, p. 450).
How to apply the theory of complexity to the izational behavior in has been little studied. In
particular, computable models of crisis and using lexity theory are virtually absent.
One needs a specific theory that link observables leading to models and simulations that can be compared to the
behavior of real cases. In other words, a middle-range theory must be found. By a middle-range theory one
understands “theories that lie between the minor but necessary working hypotheses that evolve in abundance
during day-to-day research and the all-inclusive systematic efforts to develop a unified theory that will explain
all the observed uniformities of social behavior, social organization and social change” (Merton 1968, p. 39).
System dynamics modeling is colloquially characterized by the system dynamics community as ‘theory building’.
Schwaninger and Grdsser (2008) studied whether system dynamics enables the construction of high-quality
theories. They provided strong evidence for that using a set of criteria for high-quality theories on a revelatory
case (in the sense of Yin 2009, p. 45ff). By analogy, the process of system dynamics modeling, as proposed in
major text books (Richardson and Pugh 1981; Sterman 2000; Maani and Cavana 2007) and as practiced in high
quality research work satisfies the denomination of theory building.
A recent article in the System Dynamics Review (Tu, Wang et al. 2009) applies complexity theory, in the sense
of describing the emergence of order in dynamic nonlinear systems, to develop a system dynamics model of a
the handling of a minor crisis — the Palau incident. The crisis is described and analyzed in the book “Cognition in
the Wild” (Hutchins 1995). Hutchins data collection was performed on the vessel ”Palau”, on a four-day joumey.
On the last entry to the port the ship lost its power, and all electrical systems shut down. The navigation crew
guided Palau into port despite the malfunctioning equipment. The crew had to operate on the verge of chaos,
responding continuously to navigational cues obtained by calculating the ships position manually based on
landmarks.
The Palau incident was a minor crisis, but it does share important characteristics with major crises and
emergencies. Tu et al. argue that the handling of the Palau incident followed a common set of related events,
disequilibrium—experimenting-emergence process as described by MacIntosh and MacLean (1999). This
disequilibrium-experimenting-emergence process rhymes quite well with the empirically founded principle that
successful handling of emergencies requires that the organization transitions to an Emergent organization (Dynes
and Quarantelli 1976; Turoff, Gonzalez et al. 2012).
Tu et al.’s system dynamics model can be seen as a middle-range theory for the Palau incident. The authors show
that the model renders the reference behavior of the emergency management as derived from the analysis
provided by Hutchins (1995) and satisfies the canonical verification and validation principles of best-practice
system dynamics modeling. Thus, Tu et al.’s work satisfies the criteria proposed by Schwaninger and Grésser
(2008). In fact, Tu et al.’s work satisfies also this requirement: “For a theory what is required is a model along
with a plausible account of why the model produces the behavior that it does” (Lane 2008).
Because of the relevance of Tu et al.‘s fr ion study was is extremely
important to science (Jasny, Chin et al. 2011), but ore studies in social and management sciences were
rare (Sterman 2000, p. 855ff) and probably are still rare. Our replication study, with kind help from Y a-tsai
Tseng, the corresponding author in Tu, Wang and Tseng (2009), confirmed their main findings related to the
emergence of order in dynamic nonlinear systems. However, the system dynamics model used for the results
described by Tu et al. is very complex. Further, there were several errors in the paper, including a mismatch
between the actual simulation model developed by the authors using ithink and the model diagrams in the paper,
which were drawn with Vensim. The replication study could show that the system dynamics model could be
significantly reduced in size and simplified, while still being able to render the reference behavior from the Palau
case. Further we extended Tu et al.’s work and developed a preliminary system dynamics model of the Hatlestad
Slide (Gonzalez, Bge et al. 2013).
The work by Gonzalez et al. (2013) was preliminary on two accounts. First, the reference behavior modes are
quite speculative since they were derived by generous extrapolations from the description of the Hatlestad slide
in a master thesis (Lango 2010) and a book chapter (Lango 2011). Both sources, while rich in details, did not
target a potential system dynamics simulation, so that the data therein that could be used to compose reference
behavior for crucial variables is rather incomplete. Second, the study extends a (much simplified version) of Tu
et al.’s paper (2009), which is not optimal, owing to the large differences between the Palau case (a minor
incident) and the Hatlestad slide (a major tragedy).
In this work we develop a system dynamics model from scratch based on reference behavior modes obtained by
interviewing first responders to the Hatlestad slide. The structure of the paper is as follows. First, the Hatlestad
Slide is described. Second, we present the reference behavior obtained from interviews with first responders.
Third, the structure of the system dynamics model is discussed. Fourth, a simulation and feedback analysis is
conducted. Fifth, the model’s simulation behavior is shown to agree quite well with the reference behavior
modes and a feedback analysis is conducted. Sixth, we show that the model passes a series of extreme tests.
Seventh, a sensitivity analysis is conducted. In the last section we discuss the obtained results and suggest that
the system dynamics model could be considered a rudimentary middle range theory for emergency management.
THE HATLESTAD SLIDE
Hatlestad Terrasse is a neighborhood in the city of Bergen, Norway, consisting of housing units in a rural hilly
setting. In the early morning of 14 September 2005 the cliff above the neighborhood broke apart, and a slide of
clay, mud and rock hit a row of four houses while their residents were asleep. The slide went through the ground
floors without causing the houses to collapse. People on higher floors were not harmed, and they could evacuate
without need of assistance through windows in the first floors. But the residents in the ground floors were
severely afflicted: ten people were buried and three died as a result, while seven more persons were wounded. A
total of 225 people in the Hatlestad surroundings were evacuated (Wikipedia 2007).
The disaster investigation pointed to extreme weather conditions in ination with the of the hill
behind Hatlestad Terrasse as a major cause of the landslide (Multiconsult AS 2005). For several weeks before
the disaster Westem Norway experienced record-breaking precipitation. One weather station, Bergen-Florida,
recorded the highest ever precipitation (156.5 mm/day during the fatal day of 14 September 2005) since the
establishment of the weather station in 1875 (Meteorological Institute Oslo 2005).
The Hatlestad slide was the first of a series of slides in Norway. Shortly afterward a new slide occurred in
Helebakken, Bergen. A new major slide occurred in Alesund 2008 and in Namsos 2009. The total impact of the
slides can be considered an agenda-setting crisis (Boin and Hart 2008, p. 19; Lango 2010, p. 91). The Hatlestad
tragedy and the other slides in Westem Norway were extensively covered in the media and had a strong impact
on public opinion, so that the policies for housing construction on hills became stricter. Also, preparedness
toward extreme weather changed. A fter the major slides, evacuation of citizens at risk is routinely considered in
conditions of extreme weather. Other preventative measures were implemented, so for example the city of
Bergen, which mapped out housing potentially at risk in case of extreme weather (Wikipedia 2007). Furthermore,
firefighters were given new equipment and got training for this kind of situations.
REFERENCE MODES FOR THE HATLESTAD SLIDE EMERGENCY
In order to obtain the reference modes of key variables, several interviews were carried out with first responders
that took part in the Hatlestad slide. Although these interviewees got a first-hand experience, it is important to
note that these interviews were carried out 8 years after the tragedy and therefore, they had some difficulties in
remembering all the details. Thus, at the beginning of the interview, the timeline of the Hatlestad slide was
d to help ber all the that occurred.
Initially, the invitation was sent to nine first responders from different fields (police, health, civil defense,
municipality and firefighters) to participate in the interviews. Nevertheless, only 4 experts agree to participate in
the process (see Table 1). The reasons for not taking part in the process were different: some experts argued that
they were no more working in the same position and therefore they preferred not to take part in the process.
Others argued that they did not remember much about the tragedy and they did not feel able to answer the
questions. The interviews were conducted in the autumn of 2013. All of the interviews took place at the
interviewees’ workplace.
Organization Role
Hordaland Police District, Bergen Center Emergency management, steering and coordinating
Hordaland Police District, Bergen South Emergency management, steering and coordinating
The Civil Defense, Hordaland District Supporting function
Municipality (Bergen), Section for Safety Take care of people affected by the landslide both
and Preparedness under and in the aftermath of the landslide
Table 1: Organizations of experts that took part in the interview
The interview had two main objectives: (1) to confirm that the already made hypotheses are correct and (2) to
obtain reference modes to validate that the model represents the reality. In the first part, short questions were
asked to the interviewees regarding some experienced situations to confirm some of the already set hypotheses.
These questions were related to the availability of information during the emergency, amount of data during the
emergency, understanding problems among different teams involved in the emergency, the need for
improvisation, the causes of errors committed etc. The experts were asked to evaluate from 0 to 5 (being 0 no
experience and 5 high experience) to what extent they experienced the proposed situations.
Experts stated that the following situations are the ones that they more experienced:
e Misunderstandings with other teams involved in the emergency: different stakeholders often use
different vocabulary or they have different procedures or priorities when responding to a crisis. This can
lead to mi dings and problems when making decisions about what has to be done.
¢ — Increase of workload: when non-unpredicted events occur, these may generate high-stress situations.
Beforehand established procedures may not be suitable and as a result, the workload of responders
increases significantly above normal situations.
e Need to improvise response actions: related to the previous situation, when already defined procedures
are not sufficient or suitable to respond, new response actions need to be improvised to properly
response.
e — High stress situation: every ¢ Crisis Creates stress situation and the non-unpredicted ones create even
higher stre: Hig can lead to commit errors in performing tasks and
making decisions.
In the second part of the interview, the experts were asked to draw the behavior over time graphs of key
variables, which were identified from previous work (Tu, Wang et al. 2009; Gonzalez, Bge et al. 2013). Bearing
in mind that the meaning of some variables could be confusing, the survey provided a concrete definition for
each variable. Furthermore, during the interview, the interviewer explained each variable informally using some
examples to ensure that the meaning the expert was using corresponded to the one provided by us. Below, the
analyzed key variables and the given description to each of them are explained:
e — Expected/Actual workload: when a crisis occurs the workload of responders increases above the normal
level. These variables measure the quantity of task that needs to be carried out during the emergency
and the number of tasks actually performed in each moment, respectively.
e Cognitive load: when we are doing something very familiar, we use a small part of the mental capacity.
On the other hand, when we are performing something very unfamiliar, or performing several activities
at the same time, we might need to use more mental capacity.
e Improvisations: already defined procedures and protocols are often not suitable to cope with
unpredictable situations. In these cases, improvisations have to be done, such as defining new initiatives
or improvising emergent solutions to respond to the crisis.
e Mismatch of understanding: this variable represents the difficulties that different teams (stakeholders)
in response activities may experience in understanding new decisions and actions that have been
improvised locally by other members due to the extraordinary situation.
e Quality of the decisions: when facing an ex dinary situation, the decisions often have to be taken
under stress situations and without much information. A fter the emergency, when looking back to the
taken decisions we might see that they were not the most appropriate or efficient ones. Thus, this
variable refers to how good or proper the decisions taken during the situation are.
The aim of this exercise was to gather information about how these key variables evolve during the Hatlestad
tragedy. This research was not interested in the concrete values but in the shape of the curves, that is, if the
variables increase or decrease, how much and for how long these variations remain (using relative values).
The graphs obtained vary from one expert to another in some cases. The biggest difference lies in the graphs
provided by the fourth expert since his role was quite different from the rest of the experts. He had a supporting
role that means that his role was to take care of people affected by the landslide both under the management and
in the aftermath of the landslide. So while the other actors were almost finished with their tasks the active role of
this expert had just started. Following, the graphs drawn for all the variables by the experts are presented.
fo & Expected workload
oo . —— Expected Workload A. Aetual workload
Me — Actual Workload _
RX . a
es
ooo e200 coo ov00 080 000 on00 eeco amo ano 1100 1200 iG Ge a a a ea A Ge ee ee ee
(A) (B)
: — Expected workload
— Aetual workload
cxoo cron co etm sco ono mmo on erm sooo atoo Tam”
©) eso oioo om okoo wen tem thon ae
(D)
Figure 1 Expected/A ctual Workload
The expected workload was initially higher than the actual workload owing to initial delays in the response.
Later, the actual workload was higher than the expected workload because unexpected tasks arose. The workload
was high initially and decreased along with the response and recovery activities. The graph D is different from
the others since, as explained before, the role of the fourth expert was different. He started responding to the
tragedy around 3 am. Around 6 am, he arrived to the evacuee center to help to evacuate people and provide
support. Therefore, the B graph best the ref
mode for e: d workload.
(A)
Ee
ois 0200 0300 ckd0 osm em _G790 089 ceo 1000 100 sao)
(B)
ouoo oreo 0300 akon oso auco O70) 200 100 a0m> uo 200
(C)
(D)
Figure 2 Cognitive Load
When the emergency occurred the disaster situation was new for the responders and therefore, they had to
process a high amount of information and interactions to understand the situation and make decisions about how
to respond. Once the situation was understood and the disaster started to recover, the cognitive load decreased
eventually. Similarly to the previous case, the fourth graph represents a different situation since the role of this
expert was different. Among graph A, B, C, we choose graph A as reference mode as the need to process
information and task actually increases at the beginning, therefore, the cognitive load should not peak at the
beginning, rather, increase for a while and then decrease.
(A) (B)
ea czao asnn ofeo soo soo aren akan cpm wee 11n0 1ne0 von clon u0 o4cnonco ovoo whom ohoo onuo anco ston thon
(C) (D)
Figure 3 Improvisations
In the beginning, as the situation was not previously predicted, several improvisations were needed in order to
respond effectively. In the first hour, the number of improvisations was high and eventually it decreased. As in
the previous cases, the graph D has a different behavior. In this case, the number of improvisations that need to
carry out increases eventually because they have more and more affected people to manage and provide support.
Graph A and C are similar, both expressing the idea that improvisations increases at the beginning and then
gradually reduce. We pick Graph A as reference mode.
ia De Cn GD) Ge Re Ge Ge a ee ee oxo arco c200 0190 080 os00 0700 0840 o800 atc 100 i200
(A) (B)
x00 oo 300 foo os0 osm 0700 aD 3:00 1000 suo 32.00 x00 oreo 0300 oreo 500 aso 07.00 ago 09 1000 at00 fe00
(Cc) (D)
Figure 4 Mismatch of understanding
Although dination activities are developed within the ion period to improve the coordination among
responders, stakeholders argued that initially they had some trouble to understand communications from other
the teams but eventually, after several hours working together, the quality of the mutual communication
improved. Accordingly, we choose graph A as the mode for the mi of
do Ge a ie ee a a oxo ara0 000 o4ao sn osm 0700 ono oncD 3000 anim 1200
(A) (B)
See ‘90 e200 o340 oto Os00 cG0D ore CaO Hm «OD 1109 1200
(C) (D)
Figure 5 Quality of the decisions
In the beginning the stress situation was high and therefore, stakeholders were not able to make the most
appropriate decisions. They experienced several trial and error situations. A fter some time, the stress situation
decreased and as a result, the quality of the decisions increased.
In order to develop the simulation model, the relationships between the main variables are required. We asked
the responders for the relationship of the following two pair of variables: the relationship between quality of the
decisions and cognitive load and the relationship between quality of the decisions and mismatch of
understanding.
uatty ofthe
decisions
‘ustty ofthe
decisions
Comnitve Load
(C)
Cognitive Load
ualty ofthe
deciions
Cognitive Load
(B)
‘usity of the
Cognitive toad
(D)
Figure 6 Relationship between quality of the decisions and cognitive load
In this graph the aim was to determine the relationship between the previously defined two variables: quality of
the decisions and cognitive load. However, the experts understood this graph differently. Some of them stated,
“more thinking leads to better decisions”. This means that if responders have more time to think the quality of
the decisions will be better since they will be able to analyze all the consequences that a decision may have
(graph A, Band D). Having high cognitive load refers that responders have not enough time to properly think
about the decision and therefore, the quality of the decisions will be low. On the other hand, the experts number
three stated that if the quality of the cognitive load is high then the quality of the decisions will be low and the
opposite (see graph C). However, both perspectives lie within the same basic idea: high cognitive load leads to
making low quality decisions. Therefore, we choose graph C to represent this reference mode.
usity ofthe
elsione
Mismatch of understanding
Quality ofthe
decisions
(C)
Mismatch of understanding
Quai ofthe
decisions
uty ofthe |
deesons
Mismatch of understanding
Mismatch of understanding
(D)
Figure 7 Relationship between quality of the decisions and mismatch of understanding
The target of this graph was to determine the relationship among the previously defined two variables: quality of
the decisions and mismatch of understanding. In this case, all the experts agreed on the behavior of this graph.
All the experts considered that the relationship among the ually a of the decisions and mismatch of
ding is the opposite relationship. Higher the mi ding lower the quality of the
decisions will be and the opposite. Therefore, all of the four met express a similar behavior and we can choose
either one as reference mode.
THE STRUCTURE OF EMERGENCY MANAGEMENT SYSTEM DYNAMICS MODEL
Dynamic hypothesis
Previous literature (Tu, Wang et al. 2009) and interviews with Hatlestad emergency responders suggest the
dynamics hypothesis. Tu et al. proposed that a simple core causal loop diagram with two reinforcing loops and
two balancing loops represents the handling of the Palau incident. To confirm that this hypothesis also would
apply for the case of the management of the Hatlestad emergency, we interviewed the Hatlesstad emergency
responders. .
Error corrected
8: Lear
ning
‘Mismatch
- +
+
Mutual
Understanding
Errors from
mismatch
R
Mismatch
leads to
error
ook
Work Rate
Workload +
+
R
Cognitive
load leads
Errors from
to error, a
cognitive
local innovation * pressure
SN eoaitive load
Figure 8 Causal loop diagram
An important balancing loop is the “Perfc i ” loop. When the emergency happens additional
workload is generated. The cognitive load of persons handling the emergency will increase accordingly. To
reduce the cognitive load, officers will initiate local innovations and the work rate will improve. Thus, the
workload decreases.
Two reinforcing loops relate to errors, viz. “Mismatch leads to error” and “Cognitive load leads to error”. Errors
can occur owing to the cognitive load and the mismatch in communication. Upon the onset of the emergency
happens the cognitive load increases and errors will be generated from cognitive pressure. This will further
increase the workload. Under heavy workload and high cognitive load, local innovations take place. Normal
working routines are thus disrupted and mutual understanding drops, leading to mismatch among team members.
This will breed errors from mismatch, which eventually cause additional workload.
A balancing loop, “Learning”, expresses that error correction gradually increases mutual understanding and
reduces mismatch, which leads to fewer errors.
Model sectors
Workload and cognitive load
nitive Load>
Errors rate from
mismatch
Errors rate from
cognitive pressure
QD enors
Normal work rate
Cumulative
OS — pe Work Rate
rate of local Improvement
innovations
Workload from gry,
emergency
B: ne>
Performance
adjustment
\
Improvisations
Indicated cognitive
load
Cognitive Load
hi te of
onangs rele O| <+——-Agjusstment time
cognitive load
Figure 9 Model structure—W orkload and cognitive load
The variable ‘expected workload’ represents the quantity of tasks to be completed during the emergency, which
is composed of ‘expected workload from emergency’, ‘normal work rate’ and ‘errors’ (referring to the added
workload caused by errors).
expected workload= expected workload from emergency ~normal work rate + errors
‘Workload Backlog’ is drained by ‘actual work completed ’. When ‘expected workload ’ is higher than ‘actual
work completed ’, the backlog of workload increases. ‘Workload Backlog’ generates cognitive load. In
accordance with the disequilibrium—experimenting—emergence paradigm the responders react to the disruption
induced by a high cognitive workload with improvisations. Over time (after a learning process, see below) the
improvisations improve the work rate. The variable ‘actual work completed’ increases and this reduces
‘Workload Backlog’ and ‘Cognitive Load ’. This part represents the balancing loop “Performance adjustment”.
The variable ‘expected workload from emergency ’ represents the additional workload caused by the disaster. In
the Hatlestad slide case, according to the interviews, the additional workload from is high at the
beginning and gradually decreases to zero as more and more routine tasks replace the emergent task from
disaster conditions. The variable ‘errors ’, representing the added workload from errors, include errors generated
by mismatch and errors generated by cognitive pressure.
Mismatch and Errors
When local innovations are implemented, the “Mutual ing” The “Mutual ing” is
das 0-1 scale, rep ing no mutual ing/full mutual und ding. Mismatch will be
created when “Mutual understanding” is less than 1.
Average time to
M3: Learnin
a, correct mismatch
increase of Mutual
t "
Mismatch
mismatch corrected
mismatchgenerated
ia Time to generate 6 Load=
nat mismatch Errors rate from
Time to decrease | Understanding mismatch
mutual understanding
Tg | Time to increase MU Errors rate from
B- decrease of Mutual cognitive pressure
pe Understanding
Effect of local Ee tie a i
Innovetlonon MU oO Normal work rate
load leads
fo error pax
KHesired workload
Cumulative | <Normal work rate>
Ope Work Rate
Improvement
rate of local e
innovations
Workload from grime.
emergency
Figure 10 Model structure—Mutual understanding and mismatch
Mismatch will be corrected over time. During the process of correcting officers will
with each other, and thus, the “Mutual understanding” will increase. When “Mutual understanding” rises, the
“Mismatch” will be lower than it otherwise will be. This represents the balancing loop “learning”.
SIMULATION AND FEEDBACK ANALYSIS
Stage 1: From the Ist to the 90th minute.
Figure 11 illustrates variables related to workload. The blue line with number 1 is “Normal workload”, which
remains ly at 10 When happens, addi il tasks occur, which are represented
by “Workload from emergency”. This variable is an exogenous input starting at 6 task/minute (60% of normal
workload) and gradually reducing to 0. This corresponds to the Hatlestad Slide case situation. For other cases,
“Workload from emergency” could be set differently.
The green line (3) represents the “desired workload”. When the emergency happens, “Workload from emergency”
leads to unusually high “Desired workload”. Facing high “desired workload”, responders will implement local
innovations to improve work efficiency and try to complete the work. However, there is a delay in the
of the local i , therefore, the ‘actual workload completed ’ follows the growth of
“Expected workload’ with a gap. When * actual work rate’ is lower than ‘Expected workload’ workload required,
the workload backlog increases and so does the “Cognitive Load”. Higher cognitive load leads to more errors
from cognitive load. At the same time, new local innovations reduce mutual understanding and cause errors from
mismatch. The “Desired workload” increases in the first hour mainly because the additional tasks to correct the
errors (see figure 12). The workload from emergency gradually decreases and the “actual workload completed”
increases due to the local innovations.
Workload
~4_|
a oe
r2+—~12
0
0 60 120 180 240 300 360 420 480 540 600
Time (Minute)
Nommal work rate : baserun
Workload from emergency : baserun
desired workload : baserun
actual workload completed : baserun
Figure 11 Simulation result—workload
In figure 12, we can see that the “Cognitive load” peaks at the point that “actual work completed” cross over the
“desired workload”. When actual work rate is lower than workload required, the workload backlog increases and
so does the “cognitive load’. When “Cognitive load” increases, the local innovations are initiated and work rate
is improved. The “Cumulative work rate improvement” behaves similarly as “Cognitive load”. When
innovations are implemented, mutual understanding is reduced and mismatch is increased, which causes high
“errors rate from mismatch”. Errors rate means the percentage of errors in the total work completed.
Cognitive Load Cumulative Work Rate Improvement
1 08 4
06 HF
|
" . |i
Boa
A |
~ |
02
rs t C
|
!
o L| 0 Ih
0 120 1a) 240 300 360 20 480 540 120 180 240 360 20 48040. «G00
Tine (Minute) Time (Mime)
Cognitive Load: baserun Cumulative Work Rate I id
Mutual Understanding Mismatch
1 Sse 2
075 18 > NE
[ |
i = z 4
Eos 4 Baty
/ |
!
oas LI os Lt
¥
° ol
60 120 0 300 360 420 480 S40 600 T0180 240 360
Tine (Mite) Tine (Mine)
Enns rate from mismatch Envors rate from cognitive pressure
02 02
|
os ous H
|
Boot Boo
| |
005 00s [L
|
|
ol o tf
LJ
30 120" 180 HO 6 3 70540600
240 300 360 240300360
‘Time (Minute) ‘Tine (Minute)
Figure 12 Simulation result
The “Errors rate from cognitive pressure” is highly related to “cognitive load”. Therefore, “Errors rate from
cognitive pressure” behaves in the same way as “cognitive load”. Regarding “Errors rate from mismatch”, when
“cognitive load” is high, local innovation happens and this reduces mutual understanding. As a result, the errors
generated by mismatch increases sharply at the | ing. A ination of the abor ioned two types of
errors, “errors” increase sharply at the beginning, perk around 90 minutes and then starts to decrease. These
errors add additional tasks to “desired workload”. Therefore, the “desired workload” increases in the first hour.
Errors
4
ana
Nm
2
1
0
60 120 180 240 300 360
Time (Minute)
Enors : baserun
Figure 13 Simulation result—Errors
The system behavior in the first one and a half hours is dominated by the two reinforcing loops “Cognitive load
leads to errors” and “Mismatch leads to errors”. At the same time, the balancing loop performance adjustment is
working, the “actual work completed” is increasing. However, as there is delay in the system, the “actually work
completed” is still lower than the “desired workload”.
Stage 2: From the 90th minute to the end.
Over time, the local innovations are implemented and the cumulatively the work rate is improved. The “actual
work completed” increases to the level that is higher than the “desired workload”. This leads to a reduction of
“workload backlog”. Therefore, after the first 90 minutes, the major loop that dominates the system behavior is
the balancing loop—performance adjustment. The “cognitive load” starts to reduce and as a result, the errors
generated from cognitive load are reduced. As responders work together, mutual understanding increases and
errors from mismatch also reduces. Moreover, the correction of the mismatch leads to learning, which also raises
mutual understanding. Fewer errors are generated. As a result, the “desired workload” is lower than it otherwise
could have been.
In the stage, the balancing loop performance adjustment’s effect become obvious and then the system switch
from a vicious loop (more errors generate more work, which add to workload and generate even errors) into a
virtuous loop (less workload backlog, reduces the errors, which reduces the workload further). The other
balancing loop of learning is also dominant during this stage
COMPARISON WITH REFERENCE BEHAVIOR
System dynamics model’s robustness lies in the correctness of the model structure. The model composes four
causal loop diagrams that are in accordance with the interview of the personnel handling the emergency. We
compare the reference mode provided by the informants with the model behavior.
Table 2 Comparison of model behavior with reference mode
Desired workload and actual workload completed
Workload — € Expected workload
20 A Actual workload
PS \
”
4]
E fan
H ua
|
eI) .
ul! Pm} E
I =
a 01:00 0209 05:00 04:00 05:00 06:00 07.00 O80 8:00 1001100 12-00
0 6 120 180 240 300 360 420 480 540 600
Time (Minute)
desired workload : baserun. ——?—4—3-—3- 34443.
eted: baserun
Cognitive Load
Cognitive Load
1
0.75 f Le
E os |
4 j .
} =
025 | =
5 . =
0 6 120 180 240 300 360 420 480 540 600
‘Time (Minute)
pitve Load : basen *
Improvisations
Improvisations
0.06
0.0435
E 002
0.0108 coy one clon otc oon taco clon econ coop an nhon es
-0.006 CAIZD EEC EE im
0 60 120 180 240 300 360 420 480 540 60
‘Time (Minute)
+ baserun
Mismatch of Understanding
Mismatch
m0 300 360
‘Time (Minute)
|
40 540 600
Mismatch: base
x0 0200 e200 0800 0500 0800 07:00 0ea0 oH-0) 1000 1100 120
Relationship between quality of the decisions and cognitive load
[Graph Lookup - Errors rate from cognitive pressure
uatty ofthe
| Groph Lookup - Errors rate fram mismatch
‘ual ofthe
rp) Ohput
. [a
iS)
Ties (ou
(o2eas (oa2ni8
oarat_ (canes
(0486 (C080ee
(Genes (oareor
oer? ens211
cares (003825
7 [or
decane
Input Outs
1 [0 4)
72 [io
c+ loos
cs [oos 4
if [ons
a7 014 ee
ae [oro
& pas Cognitive Load
+ [02
a La
Error rate from cognitive load
Relationship between quality of the decisions and mismatch
of understanding
neo _—
Error rate from mismatch
Mismatch of understanding
The next two sections concern extreme tests and sensitivity analysis of the model.
EXTREME TESTS
In this section, we conduct a set of extreme test to validate the model. Below is a table summarizing the extreme
tests with the detailed information on parameter change.
Simulation name
Base run parameter setting
Extreme test parameter setting
Extl No emergency
WITH LOOKUP(time, ([(0,0)-
(800,20)],(0,6),(180,3),(600,0) ))
Workload from emergency (=0)
Extreme long time to adjust
Adjustment time=10 minutes
Adjustment time=200 minutes
cognitive load
Extreme long time to time to Time to decrease mutual Time to decrease mutual
decrease mutual und di d ding=3 minutes understanding=200 minutes
Extreme long time to generate
Time to generate mismatch=10
Time to generate mismatch =200
mismatch minutes minutes
Extl No emergency
Variable change: W orkload from emergency (=0)
Expected result: Without any workload from emergency, the system will be in equilibrium condition. No
variable will change.
actual work completed desired workload
actual work completed desired workload
20 20
Jt a
7 0
5 ur au
e |] é
ul aoa 1 tt
8 8
CT eT) 30 a0 480 SAO 600 ia0 1ad 240 300 30020 e] «SHO 0
Tine (Minse} Tine (Mine)
ic completa: . + desired +
* ited: Ex] 2 Bl
Cognitive Load Mutual Understanding
Cognitive Load Mutual Understanding
1 ay 1 =F
f \
/
075 a] 075 rt
a 0
CET ETT) a0 20 a8 SAD 0 wa) Ted 240 30 0 aad BAO 0
Tine (Minze} Time (Mite)
ive Load :b Mut
fie Load :Exl. 222 —= Mute
Mismatch Errors
Mismatch Enors
2 4
15 3H -
« {
| a] i
1 - a2 t
] Ss i
{ 6
os LI 1
l N
0 0 “toh.
60 “Tao” 180" "240" 300 "360" 420 eo 70 Tao Teo Ha ho “36 0 “aa S40 coo
Time (Mint) Time (Mine)
Misaich:besenm +++ Mismatch: x1. >>> — mos :basenan Emon:Exl 2222
Figure 14 Ext1 No emergency. Model behavior:
(Base: Blue line, with number 1; Extreme test: Red line, with number 2)
Result: Model simulated results fit the expectation. Extreme test passed.
Extreme long time to adjust cognitive load
Variable change: A djustment time (=200)
Expected result: Cognitive load changes slowly, therefore, the peak will come later. Similarly, other variables
also peak later.
actual work completed desired workload
actual work completed desired workload
20 20
7 + le - 175
4 bi aU
aut 23 epp 44
EI] i fs = i
uff sat. | | 2s
8 10 i=
o 60 10 180 240 200 360 420 480 540 600 o % 120 10 m0 300 300 20 40 HO 60
Time (Mime) Tine (Mite)
.
E deste wortoed : EX2
Cognitive Load Mutual Understanding
Cognitive Load Mutual Understanding
1 = 1 sy
f Lee \
/ | |
05 07s
| : ¥
@ ae LI ea Pakbule | a
Eos} Fos Lt
| “he \
025 “i 0.25
a °
Go 20 Te 24 30 a6) RDO G0 0 120 Te) mad a0) so) DW) HO GN
Tine (Mingo) Time (Mine)
tive Load Mut
tive Load :E Mute
Mismatch Errors
Mismatch Errors
2 4
15 aH +7 SESS
4 aL A {
2 - HT i
Ba g2
! i |
os LI 1H
0 sft || old Pett ad |
0 20 ie 240 300 360 420 400 S40 G00 oo 120 ~<a 240 300 360 420 480 40 G00
Time (Mint) Tie (Mint)
Mismatch basen Mismatch: x2 Emors:baenn +++ Ess?
Figure 15 Ext2 Extreme long time to adjust cognitive load. Model behavior: (Base: Blue line, with number 1; Extreme
test: Red line, with number 2)
Result: The desired workload doesn’t increase at the beginning. This is because in the baserun, the desired
workload increases at the beginning because the increase of errors. In the extreme test, errors doesn’t go up
quickly, therefore, the desired workload doesn’t increase. Extreme test passed.
Extreme long time to time to decrease mutual understanding
Variable change: time to decrease mutual understanding (=200)
Expected result:: This variable means how fast mutual understanding are reduced when local innovations are
implemented. In the baserun, when local innovations happen, in 3 minutes, mutual understanding will be reduced.
In this extreme text, the mutual understanding will be reduced in 200 minutes. Therefore, “Mutual understanding”
and “Mismatch” will not change as much as baserun. “Errors” will be smaller and thus the “desired workload”
will not increase. “Cognitive load” will not be as high and “actual work completed” will not increase as quickly
actual work completed desired workload
actual work completed desired workload
20 20
1 = 5
i ic = kK
eu | a5 Bl
i be a
u ns
8 0 =e
G02 180 24030060 70 40 G00 %0 120 Ted 20 300 30 DB) HO ot
Time (Mite) Time (Mite)
. desired
Bg 22> 53
Cognitive Load Mutual Understanding
Cognitive Load Mutual Understanding
1 7 1
f Lat \
iM ao |
075 =P
| Ras |
s \
4 \
— 0.25
0 o
GO 20 Te 240 300 a6 OAD CHAD G0 0 1 Te) Md 300 a0 QD WD HO GN
Tine (Mingo) Time (Mite)
rtive Load :b * * Mut
ve Load :B r Mute
Mismatch Errors
Mismatch Errors
2 4
15 = mr +7
a | 2? ||
z | é, (1
Ea g2
| fs Ie
os LI 1 as
old = fj. 0 te-bsl |
0 “Tao” 180 240" 300 360 420. 400 S40 G00 oo 120 ~<Tao 240 300 360 a2 480 S40 G00
Time (Mint) Tie (Min)
Mismatch basen Mismatch: x3. —2>—2—2— mos basen Emon:EX} 222 >
Figure 16 Ext3 Extreme long time to time to decrease mutual understanding. Model behavior: (Base: Blue line, with
number 1; Extreme test: Red line, with number 2)
Result: Model stimulated results fit the expectation. Extreme test passed
Extreme long time to generate mismatch
Variable change: time to generate mismatch (=200)
Expected result: “Mismatch” changes slower than baserun and thus “Errors” will be less. Less learning happens
and “mutual understanding” will be less than baserun. With less “Errors”, the “desired workload” will not
increase. The “actual work completed” will be slower than the baserun
actual work completed desired workload
actual work completed desired workload
20 20
1 = 5
3 mec | | L au
any a5 L |
Ee yi) Le & TPT RIS RER
ni = 25
=|
8 10 “cE
G2) 180 240 300 360 20 ae SHO. G00 %0 120 Ted 20 300 se) DB) HOON
Time (Mite) Time (Mite)
. desired
Ed 22> 2 Bd
Cognitive Load Mutual Understanding
Cognitive Load Mutual Understanding
1 = 1 a
f Le \
iM ao |
075 075 a
| 2
3 | 3
Eos Eas th
aaa s \
| \ \
0.25 — 0.25
0 o _
GO 20 Te 240 300 a6 OAD CHAD G0 0 1 Te) Md 500 a0 OD WD HO GUO
Tine (Mingo) Time (Mite)
* * Mut
Cogutie Load: Ext > Mute
Mismatch Errors
Mismatch Errors
2 4
15 + aH +7
| % i
i | é H
Fu 2
as | 3 a
i
«| lt
os LI iu
i ]| ~
= }
ol be} ¢-* Spe 0 ‘ “RELL
0120 180 240 300 360 420 480 S40 600 oo 120 ~<Tao 240 300 360 az 480 S40 G00
Time (Mite) Tie (Min)
Mismatch: basen —2—2—3- Mismatch: Exd mos basen Emon:Exd 2222
Figure 17 Ext4 Extreme long time to generate mismatch.
Model behavior: (Base: Blue line, with number 1; Extreme test: Red line, with number 2)
Result: Model stimulated results fit the expectation. Extreme test passed.
SENSITIVITY TESTS
In the section below, we conduct sensitivity tests to validate the model. Two kinds of sensitivity tests are
performed. For the constants, we will test model behavior of a certain range of the constants. The range is
decided based on the rational setting of the constant. For lookup functions, we will test the different shape of the
look up, convex, linear and concave.
ST1 Time to decrease mutual understanding
Base run: Time to decrease mutual understanding = 3
Sensitivity test: Range: 1-5; Distribution: Random Uniform; Runs: 200;
Simulation behavior:
actual work completed desired workload
3H oT
basen basen
50% 7596 95% 100% 50% 75% 05% 100%
actual work completed desired wordoed
20 20
7 / 15 {fr
4 5
rn ns
4 0 150 3 450 600 af o 150 300 450 600
Time (Minute) Time (Mite)
Cognitive Load Mutual Understanding
HT or
basen baseran
50% 75% I 95% 100% 50% 75% N05 % 100%
Cogutve Load Mutual Understaneing
1 1 =
05 | 05
025 025
s 0 150 300 450 600 e o 150, 300 450
Time (Mints) Time (Mint)
Mismatch Errors
SH St
basen basen
50% 7596 95% 100% 50% 75% N05 100%
Mismatch Ems
4 | | a
3 3 [
2 2 |
1 1 ]
& ° 150 300 450 600
Tine (Mint)
Figure 18 ST 1: Time to decrease mutual understanding
The “actual work completed” and the “desired workload” do not change much. However, “Mutual
Understanding” and “Mismatch” are quite sensitive to “time to decrease mutual understanding”. This is because
“time to decrease mutual understanding” is directly related to “Mutual Understanding”. When “time to decrease
mutual understanding” is long, the “Mutual Understanding” changes slowly, and when “time to decrease mutual
understanding” is short, the “Mutual Understanding” changes fast. Yet, the “Mutual Understanding” is within the
range of 0-1. As “Mismatch” is highly related to “Mutual Understanding”, “Mismatch” is also sensitive to “time
to decrease mutual understanding”. “Errors” are partly caused by mismatch; therefore, “errors” also show some
sensitivity.
Numerical sensitivity is observed in “Mutual Understanding”, “Mismatch”, and “Errors”. But no pattern
sensitivity found. The numerical sensitivity is reasonable regarding the structure of the system.
Result: Pass
ST2- Adjustment time
Base mun: Adjustment time = 10
Sensitivity test: Range: 8-12; Distribution: Random Uniform; Runs: 200;
Simulation behavior:
actual work completed desired workload
2 7
basen basen
50% 7596 N95 100% 50% 75% 05 100%
actual work completed desired workload
20 20
7 / 175 a
u 5
" | ns
8 150 300 450 600 10 150 300 450 600
Time (Mints) Time (Mint)
Cognitive Load Mutual Understanding
siz 3
basen basen
50% 75% 959% 100% 50% 75% NN 05% 100% I
Cogitve Load Mutual Understanding
1 1
es f— “It
05 05 V
025 | 025
® 0 150 300 450 600 ° 0 150 300 450 600
Time (Mingo) Tine (Mint)
Mismatch Errors
7 ST
basen basen
50% 7596 95% 100% 50% 75% N05 100%
Mismatch Ems
2 4
15 7 3 T
. | |
0s ] 1 1
® 150 300 450 600 s 150 300 450 600
Tine (Mize) Time (Mine)
Figure 19 ST2- Adjustment time
The “Adjustment time” is the time to change “cognitive load”. Sensitivity test shows that the “adjustment” has
little impact on “actual work completed”, “desired workload”, “Mutual Understanding”, “Mismatch” and
“Errors”.
Result: Pass
ST3- Average time to correct mismatch
Base mun: Average time to correct mismatch = 30
Sensitivity test: Range: 20-40; Distribution: Random Uniform; Runs: 200;
Simulation behavior:
actual work completed desired workload
Sey oT
dasenan basen
50% 7596 95% 100% 50% 75% 05% 100%
actual work completed desired wordoed
20 20
7 / 15 f~
4 = 5 —
n 2s Ra
4 0 150 3 450 600 105 150 300 450 600
Time (Minute) Time (Mite)
Cognitive Load Mutual Understanding
Sey oF
basen baseran
50% 75% I 95% 100% 50% 75% N05 % 100%
Cogitve Load Mutual Understanaing
1 1
vs f = a
0s | Sax os Ye
025 025 LW
s 0 150 300 450 600 e o 150, 300 450 0
Time (Mints) Time (Mint)
Mismatch Errors
Sey ST
basen basen
50% 7596 95% 100% 50% 75% N05 100%
Mismatch Ems
4 a
3 3 ‘a
|
1
0 ] "
150 300 450 00 150 300 60 00
Tine (Minute) Tine (Mint)
Figure 20 ST3- Time to decrease mutual understanding
The “average time to correct mismatch” has little impact on the “actual work completed”, “desired workload”,
“Cognitive load”, and “Mutual Understanding”. However, “Mismatch” is quite sensitive to “average time to
correct mismatch”. This is because “time to decrease mutual understanding” is directly related to “Mismatch”.
When the “average time to correct mismatch” is long, the “Mismatch” decrease slowly, and when “time to
decrease mutual understanding” is short, the “Mismatch” changes However, there is no pattern change of
“Mismatch”. As “Errors” are partly caused by mismatch, “errors” also show some sensitivity.
Numerical sensitivity is observed in “Mismatch” and “Errors”. But no pattern sensitivity found. The numerical
sensitivity is reasonable regarding the structure of the system.
Result: Pass
ST4- time to generate mismatch
Base mun: Time to generate mismatch = 10
Sensitivity test: Range: 5-15; Distribution: Random Uniform; Runs: 200;
Simulation behavior:
actual work completed desired workload
STE STE
basenn basen
75% 95% 00% 50% 75% N95 100%
actual work completed dlsied worldoad
2 20
ae
/ "of
al 7 NK
° 750 300 750 0 ° 150 300 50 600
Time (Mite) Ti (Minas)
Cognitive Load Mutual Understanding
ou STE
50% 75% 95° NN 100% 50% 75% NN959 00%
Cognitive Load Muna Understanding
1 1 =
07s f 075
05 05
025 | 025
0 0 rs
0 150 300 450 600] 0 150 300 450 600
‘Tine (Minute) Time (Mints)
Mismatch Errors
STE Str
50% 75% NS 100% 5% I 95° 100%
Mismatch
Misma
2
+ |
mul
I XY
450 600] 150
300
‘Time (Minute)
300
Tine (Minute)
Figure 21 ST4- time to generate mismatch
The “time to generate mismatch” has little impact on the “actual work completed”, “desired workload”, and
“Cognitive load”. However, “Mutual Understanding” and “Mismatch” are quite sensitive to “time to generate
mismatch”. This is because “time to generate mismatch” directly affects “mismatch generated”. When the “time
to generate mi ” is long, the “mi: ” is small. “Mi: ‘h” will be smaller than it otherwise
would be. When “Mismatch” is small, less “mismatch corrected” occur, and the “Mutual Understanding” is
smaller than it otherwise would be.
Numerical sensitivity is observed in “Mi ” and “Mutual U ding”. But no pattern sensitivity found.
The numerical sensitivity is reasonable regarding the structure of the system.
Result: Pass
ST5- time to increase MU (mutual understanding)
Base run: Time to increase MU =30
Sensitivity test: Range: 20-40; Distribution: Random Uniform; Runs: 200;
Simulation behavior:
actual work completed desired workload
oS 315
basen Dasenun
50% 750 9596 100% 503% 75% NN 95% IN 10005
aca work completed deste workload
20 20
0 7 175 Fa
uu | 15
125
a 0 150 300 450 600 10 150 300 600
Tine (Minute) ‘Time (Mime)
Cognitive Load Mutual Understanding
35 315
basen
50% 750 NN 959 100% 50% 75% 95% BN 100%% I
Cognitive Load Muna Understanding
1 1
05 | 05
025 025
0 t °
° 150 300 80 600 0 150 300 150 600
Tine (Mite) ‘Time (Mime)
Mismatch Errors
315 315
baserun basen
50% 759 95° 100% 50% 75% 959% 100%
Mismatch Enos
4
3
2 [
1 f tc
600 ) 150 300 450 6
‘Time (Mine)
Figure 22 ST5- time to increase MU
The “time to increase MU” has little impact on the “actual work completed”, “desired workload”, “Cognitive
load”, and “Errors”. However, “Mutual Und ding” and “Mi: h” show a little sensitivity to “time to
increase MU”. This is because “time to increase MU” directly affects “Mismatch”. When the “time to increase
MU” is long, the “Mutual Understanding” increases slowly. “Mutual Understanding” will be smaller than it
would be. As “Mutual Understanding” directly affects “Mismatch”, “Mismatch” shows similar
Numerical sensitivity is observed in “Mi ” and “Mutual U
The numerical sensitivity is reasonable regarding the structure of the system.
Result: Pass
”’, But no pattern sensitivity found.
S6- Effect of local innovation on MU
This sensitivity test is to test the lookup function that lookup “Cumulative W ork Rate Improvement” (generated
by local innovation) to determine Mutual understanding. The lookup function in the base mun is an S-shape curve,
we test what will happen if the look up function is set up as linear and a convex curve.
Baserun S61 S62
a
——— _, Lge
see lai
abil
Simulation behavior (Baserun: blue line, No. 1; S61: red line, No. 2; S62: green line, No. 3)
actual work completed desired workload
20 20
Josie}
v {a 178
P trol "i
4 5 -
| a
ft.
u =| | 2s 3
in|
8 10
o # 120 180 249 300 360 420 480 540 600 0 60 120 180 240 300 360 420 480 510 600
Cognitive Load Mutual Understanding
foa Pre yo4 t = |
\ of ete) Ae
075 a o7s Fy CORES oe
La 2
os L os Lh bor
~| .
ee
va HEEL |
o °
o 60 120 180 240 300 360 420 490 540 600 0 60 120 180 290 300 360 420 490 540 600
Mismatch Errors
ai 4
1s is ib 3 arr AL
1 Se 2
if a r Nee
iF lh ~ mS tah |,
os 1 =
ih fa, /
° Dee 0
3 120 1s) 340 G00 Ge) Dede | «OSSD 240 «00 360 «NDB HOD
Figure 23 S6- Effect of local innovation on MU
Analysis:
Change in the “Effect of local innovation on MU” have little impact on “actual work completed”, “desired
workload”, “Cognitive Load”. “Mutual Understanding”, “Mismatch” and “Errors” show some sensitivity in the
later stage of the behavior.
The sensitivity displayed in the model behavior is reasonable.
Result: Pass
S7- Indicated cognitive load
This sensitivity test is to test the lookup function that lookup “work load backlog” to determine cognitive load.
The lookup function in the base run is an S-shape curve, we test what will happen if the look up function is set
up as linear and a convex curve.
Baserun S71 $72
je ee) |
i
Simulation behavior (Baserun: blue line, No. 1; S71: red line, No. 2; S72: green line, No. 3)
actual work completed desired workload
P29 Ty desired workload 7
20
feo
7 = <I
iy aoe ue
4 fi By Taal
5
u / ea
ile | 125
8
o 6 120 180 280 300 360 320 480 §40 600) 10
o 6 20 180 240 300 360 420 480 540
Cognitive Load Mutual Understanding
- Cognitive Load. Tg 7]
1 aa
i 1s
=|
hel 1 = 3
1 Le
i 05 || Lebeb beter T
i
°
0 o 60 120 is 240 300 360 420 480 S40 600
o 6 120 180 240 300 360 320 480 540 600
Mismatch Errors
2 4
1s 3 :
(| PSs tls!
4 aco hs
1 ai
- : LU Ne
old Ss bedacd. old
0 120 18) 240 300 360 420 480 S40 600 120 «180 240 300 360 600
Figure 24 S7- Indicated cognitive load
Analysis:
Change in the “Indicated cognitive load” have little impact on “actual work completed”, “desired workload”,
“Cognitive Load”, “Mutual Understanding”, “Mismatch” and “Errors” show little sensitivity. The sensitivity
displayed in the model behavior is reasonable.
Result: Pass
$8- Errors rate from mismatch
This sensitivity test is to test the lookup function that lookup “mismatch” to determine the errors rate from
mismatch. The lookup function in the base run is an S-shape curve, we test what will happen if the look up
function is set up as linear and a convex curve.
Baserun S81 $82
wae la
Simulation behavior (Baserun: blue line, No. 1; $81: red line, No. 2; $82: green line, No. 3)
actual work completed desired workload
[20 y 20 |
Joetesn|
7 7s Fi
ws
“4 By 1s
ofa | Ry
u re 125 am
pet
I
8 10 CRRE
o 60 120 180 24 300 360 420 480 510 600 0 6 12) 189 240 300 360 420 480 540 600
Cognitive Load
1 Fa]
o7s
ihe
os os
Soe a
02s | Pabst 33
o °
0 60 120 180 249 300 360 420 480 $40 400 0 60 120 190 240 300 360 420 480 $40 600
Mismatch Errors
tL, +, 4
15 a 3 aI
“ Pe F
1 2
os | 1
0 ° ]
o 60 120 180 240 300 360
600 6 120 «180 240 «(300 360
Figure 25
Change in the “Errors rate from mismatch” have little impact on “actual work completed”, “desired workload”,
“Cognitive Load”, “Mutual Understanding”, “Mismatch” and “Errors” show little sensitivity. The sensitivity
displayed in the model behavior is reasonable.
Result: Pass
$9- Errors rate from cognitive pressure
This sensitivity test is to test the lookup function that lookup “cognitive pressure” to determine the errors rate
from cognitive pressure. The lookup function in the base run is an S-shape curve, we test what will happen if the
look up function is set up as linear and a convex curve.
Baserun S91 S92
sgn
a eal aa
ie
a ra
Simulation behavior (Baserun: blue line, No. 1; S71: red line, No. 2; S72: green line, No. 3)
actual work completed desired workload
20 20
175 is
5
1s is
Age
8 10 ie
0 6 12) 189 240 300 360 420 480 540 600 0 6 120 180 240 300 360 420 490 $40 600
Cognitive Load Mutual Understanding
1 aT
ia yy é
0.75 1% tal
nce ° a
bs Shae - [inake|
FT
02s [Ae 02s
° o
o 6 129 187 249 300 360 420 480 340 600 o 6 120 180 240 300 360 420 480 $40 600
Mismatch Errors
2 4
1s a
“ ae r ata. |
E | SRS P| "1
05 Sy
SI N
* an
° =| Pla |
o” 6 120 18) 240 300 360 420 480 0 600 0 180 240 500 360 «<420«—480:«H40.~—600
Figure 26 S9- Errors rate from cognitive pressure
Analysis:
Change in the “Errors rate from cognitive pressure” have little impact on “actual work completed”, “desired
workload”, “Cognitive Load”, “Mutual Understanding”, “Mismatch” and “Errors” show little sensitivity. The
sensitivity displayed in the model behavior is reasonable.
Result: Pass
CONCLUSION AND DISCUSSION
The agreement of the simulated behavior with the reference behavior and the fact that the simulation model
passes a rich series of tests encourages us to suggest that the system dynamics model proposed in this paper
a mdi iddl ge theory for the transition from di ization to self- ization in
emergencies. We suggest that this rudimentary theory is worth exploring as a starting point to gain more, and
much needed und of the of ies. To establish a strong link to
working hypothesis used by practitioners and a unified grand theoty of management of emergencies more and
better data is needed.
Forrester (1980) argued that models in social science must use all three kinds of existing data, viz., data stored
mentally in people's heads (mental data), data stored descriptively in writing (written data), and data available
numerically (numerical data). The numerical data are a tiny fraction of what is found in written form, which
again is tiny compared to what people have in their heads. Excluding essential mental data would be tantamount
to trying to manage a business while ignoring perceptions, beliefs, sentiments, and all other key data upon which
decisions must be made.
The huge amount of mental data on emergency preparedness and response owned by practitioners is largely not
available for scientists. Instead, the less abundant written data, and the even less abundant numerical data shape
most of the current research on emergencies. By showing that even modest knowledge about the reference
behavior of soft variables can facilitate theory building, this paper aspires to motivate practitioners to share more
of their mental data and to inspire researchers to include direct questions to the practitioners about data series,
that is, reference behavior for even the softest of the variables in forensic studies of emergencies. In particular,
the power of Delphi techniques should be used to obtain such reference behavior.
ACKNOWLEDGMENTS
We are very grateful for valuable comments and suggestions from Starr Roxanne Hiltz, Murray Turoff, Geir Bae,
John Einar Johansen and Shengxin Chen.
REFERENCES
Boin, A. and P. t. Hart (2008). Public Leadership in Times of Crisis: Mission Impossible? Crisis management. A.
Boin. Los Angeles, Sage: 1-15.
Byme, D. S. (1998). Complexity theory and the social sciences : An introduction. London, Routledge.
Dynes, R. R. and E. L. Quarantelli (1976). Organizational Communications and decision Making in Crises.
Newark, DE, USA, Report Series 17, University of Delaware, Disaster Research Center.
Forrester, J. W. (1980). "Information Sources for Modeling the National Economy." Journal of the American
Statistical Association 75(371): 555-566.
Gilpin, D. R. and P. J. Murphy (2008). Crisis management in a complex world. Oxford, Oxford University Press.
Gonzalez, J. J., G. Bae, et al. (2013). A System Dynamics Model of the 2005 Hatlestad Slide Emergency
Management 10th Intemational Conference on Information Systems for Crisis Response and Management.
Baden-Baden, Germany.
Hutchins, E. (1995). Cognition in the Wild. Cambridge, Mass., MIT Press.
Jasny, B. R., G. Chin, et al. (2011). "Again, and Again, and Again ...." Science 334(6060): 1225.
Lane, D. C. (2008). "Formal theory building for the avalanche game: Explaining counter-intuitive behaviour of a
complex system using geometrical and human behavioural/physiological effects." Systems Research and
Behavioral Science 25: 521-542.
Lango, P. (2010). Samordning i krise eller krise i samordning? Bergen, University of Bergen. Master thesis.
Lango, P. (2011). Hatlestad-raset. Organisering, samfunnssikkerhet og krisehandtering. A. L. Fimreite, P. Lango,
P. Leegreid and L. H. Rykkja. Oslo, Universitetsforlaget: 216-234.
Maani, K. E. and R. Y. Cavana (2007). Systems thinking, system dynamics : Managing change and complexity.
North Shore, N.Z., Pearson/Prentice Hall.
Macintosh, R. and D. Maclean (1999). "Conditi issipative structures approach to
transformation." Strategic Management Journal me 297- 316.
Merton, R. K. (1968). Social Theory and Social Structure. New Y ork, Free Press.
Meteorological Institute Oslo (2005). Varsling av stormer og ekstremt veer, 18/2005. Varsling av stormer og
ekstremt veer. Oslo, Norway.
Multiconsult AS (2005). Notat G-02. Rasrisiko etter rekordnedbor 14. ber. Hatlestad. ili hold
Arsaker til skredet. Bergen
Palttala, P., C. Boano, et al. (2012). "Communication Gaps in Disaster Management: Perceptions by Experts
from Governmental and Non-Governmental Organizations." Journal of Contingencies and Crisis Management
20(1): 2-12.
Richardson, G. P. and A. L. Pugh (1981). Introduction to system dynamics modeling. Portland, Or., Productivity
Press.
Schwaninger, M. and S. Grdsser (2008). "System Dynamics as model-based theory building." Systems Research
and Behavioral Science 25: 447-465.
Simpson, N. C. and P. G. Hancock (2009). "Fifty years of operational research and emergency response."
ournal of the Operational Research Society 60(Supplement 1): 126-139.
Sterman, J. D. (2000). Business dynamics : Systems thinking and modeling for a complex world. Boston, Irwin
McGraw-Hill.
Tu, Y.-m., W.-y. Wang, et al. (2009). "The essence of transformation in a self-organizing team." System
Dynamics Review 25(2): 135-159.
Turoff, M., J. J. Gonzalez, et al. (2012). Modeling organizational behavior in emergencies. NOKOBIT, Bodo,
Norway, Akademika Forlag.
Wikipedia (2007). “Hatlestad Slide, http://en.wikipedia.org/wiki/Hatlestad_Slide." Retrieved March 16, 2014,
from http://en.wikipedia.org/wiki/Hatlestad_Slide.
Yin, R. K. (2009). Case study research : design and methods. Thousand Oaks, Calif., Sage.
