Using social network analysis as a tool to create and compare mental models

(1)

Using social network analysis as a tool to

create and compare mental models

O

VE

J

ANSSON

Linköping University

Supervisor: Jonson, Carl-Oscar

ISRN: LIU-IDA/KOGVET-A--15/004--SE

June 10, 2015

Abstract

The field of social network analysis has expanded from the field of social science to the fields of human factors and ergonomics. There is a theory that suggest that one can use the social network methods and create an information network which describes the network from an information sharing perspective and and there are also theories which describes how social network analysis can be used study cognitive maps (mental models). This thesis touches both of these subjects in an attempt to investigate how social network analysis can be used together with real-time information as a data source to investigate the cognitive maps of individuals and comparing these maps with an organisations expected structure based on protocols. The study conducted showed that it was indeed possible to change the social network analysis method into an information based network which explains the origin of a mental model and to study information be-haviour, in a network, but there are still variables which needs to be studied further (e.g. failed information sharing and temporal aspects of information sharing).

(2)

Acknowledgement

First and foremost I would like to thank my supervisor Carl-Oscar Jonson at Centre for Teaching & Research in Disaster Medicine and Traumatology for his unwavering support and and feedback during this thesis. I would also like to thank Erik Prytz Lector at Linköping University for introducing me to the social network analysis and more or less allowing me to conduct this study. My third and final acknowledgement goes to Emma Melander, a fellow student at the Cognitive Science master program, who took time to teach me about graphical design. Although I did not have time to learn anything about graphical design, but without her input the figures in this report would have looked terrible.

(3)

List of Figures

1 Basic graph illustration . . . 5

2 Graph hierarchy . . . 5

3 Reciprocity and transitivity . . . 6

4 Cliques and heterophilic role . . . 7

5 Classic social network strategies . . . 14

6 Social network graph after team improvement . . . 14

7 Simulated accident map . . . 18

8 Ambulance incident commander - nofilter . . . 25

9 Medical incident commander - nofilter . . . 27

10 joined - nofilter . . . 30

11 filter medical incident . . . 33

12 filter triage . . . 35

13 filter joined . . . 37

14 Organisational structure for rescue service . . . 40

15 Organisational structure for the police . . . 42

16 Organisational structure - healthcare . . . 44

(6)

1 Using social network analysis as a tool to

cre-ate and compare mental models

In the case of a major incident (e.g. a chemical accident) it is important that the joint incident command is working as intended and that all actors at the incident site has the correct information on which to base their decision. However, it appears to be certain flaws when it comes to communication be-tween the different organisations. The Swedish Civil Contingencies Agency (MSB) have released an report which points out that in there have been flaws in communication and situational awareness at several major inci-dents. One problem is that it is quite difficult to study the communication flow and with that it is also difficult to understand where the problem lies. It could be that there is to much information passing through one person which makes it hard to keep track of where information should be passed on to, or maybe there is simply no connection between the actors which makes it impossible to share information. Either way, if the communication is lacking, the situational awareness might change between the different ac-tors and incorrect decision might be made. To ensure optimal performance in a system, the different actors needs to follow the organisational protocols, but how can one study these factors? Interviews and observation might be a good candidate, but due to the high complexity it might be difficult to get a correct results merely from those two methods. One suggestion is that the social network analysis can be used to not only study social couplings, but also information flow. As of late, there appears to be an increase in the number of studies which are using techniques from social network theories. A technique which earlier was used in social studies have now expanded from social sciences into research fields like human factors and ergonomics. The study which this master thesis is based on takes on a slightly dif-ferent approach than earlier studies from the field of social network analysis (SNA). This study is an attempt to create a social network which represents a group of cognitive maps (mental models) and comparing these maps with an "expected map", and does so by studying real-time communication between actors in a simulation of a medical catastrophe.

In the field of emergency management there are certain rules and reg-ulation for command and control. These protocols are to be followed. If one fails to follow the protocols, the implication of incorrect information, incorrect situation awareness and wrong decisions could be severe, or in some cases, fatal. In the situation which was studied for this thesis, there

(7)

is an organisational hierarchy where the role of one person significantly differ from another (e.g. the ambulance incident commander’s role differs from the medical incident commander’s). But can one be sure that the indi-viduals in an organisation have the same mental model over the expected roles and interaction as the head of the organisation? One might argue that observation and interviews goes a long way in answering these question, but how does one make sense of the data? Deriving an individual’s role in an organisation might not be as easy as it seems. We hypothesise that one way to understand the mental models of an actor in an organisation is to study the actors behaviour using real-time data. This data is then used to derive the information flow and information demand between the different actors in an organisation. By using real-time data as a source of the mental model should allow a close fit to reality since it is actual behaviour which is studied. Using social network analysis as a method allows for a more ob-jective analysis which can be compared to the protocols of an organisation. Using real-time data means that the dataset will be exceptionally large and that is in itself a motivation for this study. If the Social Network Analysis is a method which is applicable on this kind of data, it might be a method to develop for further use in similar settings.

This thesis is an attempt to create a new method for studying organi-sational structure and interaction between different actors. By studying the information flow and information demand between different actors it might be possible to derive mental models which the individuals in the network are using to complete the task which is at hand. The model might then be able to be compared to the model expected by the organisation to see if they are close to identical or vastly different. In a way one can say that the method which is explored in this study is able to do two things:

1. Exploring mental models

2. Comparing the mental models with an "expected model"

This thesis takes social network analysis to another level; the main goal is to investigate the possibility to create social networks based on real-time communication data, and further more, using these network as a representation of a mental model which can be compared to the model which is expected to be used by the organisation. The hypothesis of this study is that a method, which is derived from the Social Network Analysis-method, is a valid tool to investigate an individuals mental model and to compare this model to an expected organisational structure. There

(8)

are indeed various ways of comparing an individuals action to expected behaviour, but by using a model similar to that of SNA one is able to create larger network in such a high resolution that one can follow important information passing through one node in the network to another. SNA was not created for a purpose such as the one which is explored in this thesis, but with some changes in the methodology the method might be seen as valid and might be a tool with great potential to explore mental models.

1.1 Research question

The following research questions were explored in this study

1. Is real-time communication a valid data source which can be used to create a social network?

2. Is it possible to use the social network approach to study behaviour and/or to explore mental models?

3. If 2: Is it possible to compare the created network with an organisa-tions expected model/structure and if so; is there a difference between the expected and the real model?

The answer to question 1 is a self assessment and will be discussed in section 5, while question 2 and 3 will be answered through data analysis. If it’s possible to create a network which represents mental models, this model will be compared to the expected behaviour from the organisation. The third research question might at first sight not seem to be connected to the first two, however, if one can create mental models through social network analysis, this model should be comparable to mental models of other individuals or, as in this case organisational expected models. If not comparable, the mental models which are created through this method would seem to lack pragmatic function in the current context.

1.2 Limitation

Although the data used in this study were massive, there was some missing data which forces a limitation on the study. The limitation of this study is that not all actors in the simulation were equipped with microphones. To ensure that the transcription of the data were as accurate as possible, only the actors with a microphone were included in the study and thus only partial networks will be created.

(9)

2 Background theory

2.1 Social network

The concept of networks is a defining paradigm in the current era (Kil-duff and Tsai, 2003) and is common in many different scientific fields (e.g. physics, and biology). By using networks as an analysis approach one is able to study the interactions of one individual component in a larger system (Kilduff and Tsai, 2003). Social networks can be used to analyse different interactions and their effect between component and in some cases the outcome of these interactions. For example; analysis of social networks has been able to predict the outcome of a strike by factory workers (Kilduff and Tsai, 2003), as well as spreading of disease (Wasserman and Faust, 1994) or analysing connections between internet users (Adamic et al., 2003)

Kilduff and Tsai (2003) Describes social networks as a field which im-ports theories from other scientific fields, most noteworthy would be math-ematical theories and social psychology theories. These theories has been molded and shaped into social networks theories which are heading into two different directions: 1) Heterophily theory which aim to make predictions of how actors’ whom are outside of the social group (explained in section 2.1.3) can access knowledge and resources from the group, and 2) Structural role theory which is about structural equivalence, structural cohesion and role equivalence. This theory tries to make prediction about how actors influence each other’s attitudes and behaviors (explained in section 2.1.4).

2.1.1 Mathematical approach

Figure 1 below illustrates different types of graphs which can represent a social network. The connections (edges) in a social network can vary from fully disconnected to fully connected (Kilduff and Tsai, 2003), and the arrows in the graph illustrates the connection between nodes (individuals) in the social network. A graph with no or few edges can imply that the communication or information sharing is impaired in a system, and a graph with a higher degree of ’connectioness’ implies a network with high information and resource sharing.

(10)

Figure 1: The network to the left represents a fully disconnected network with a degree of 0. The network to the right represents a network between family members with a degree of 5.

Graphs can also vary in levels of hierarchy (Kilduff and Tsai, 2003), and can vary from being fully hierarchical to having no hierarchy at all. In a fully hierarchical network the person at the top of the graph has a higher influence in the decision making process. Figure 2 describes a fully hierarchical network as well as a network where there is no clear hierarchy.

Figure 2: The network to the left represents a network with a clear hierarchy. The network to the right represents a network with no clear hierarchy. The arrows indicates who is "above" who.

Graph efficiency is a measurement which represents the minimum amount of links in the network while still not dividing the network in different sections. For example, one does not want to limit the flow in a social net-work by isolating different actors in a system, but too many connections in a network makes the network slower and makes it more expensive to

(11)

maintain (Kilduff and Tsai, 2003). One could say that in social network graphs, there is a tradeoff between resilience and efficiency since a resilient network demands more connections between its components.

2.1.2 Social psychology approach

Social networks borrow many different theoretical ideas from social psy-chology (Kilduff and Tsai, 2003). The balance theory. centers around how people balance their relationship to others in a social network and can be summarized as follow (Kilduff and Tsai, 2003):

1. Individuals prefer balanced relationships. The friendship should be mutual and one person’s friends should be friends with each other. For a balance there is a demand for reciprocity and transitivity (see Figure 3)

2. People prefer to interact with others whom they share some bond with (e.g. gender, ethnicity, workplace)

3. People in an unbalanced relationship have a feeling of discomfort. 4. People try to change unbalanced relationship to balanced

relation-ships, either by changing attitudes or breaking the relationship. Thus, a completely balanced network would lead to a fully connected graph. The balance theory makes the creation of cliques a fact (Kilduff and Tsai, 2003) (a clique is group of people in a network whose liking of each other is greater than their average liking of other members in a group).

Figure 3: The network to the left shows reciprocity (there is a mutual connection). The network to the right is an illustration of transitivity.

Social comparison theory tries to explain why people tend to choose to interact with certain people in the first place. According to Kilduff and

(12)

Tsai (2003), people are drawn into friendships with people who are similar to themselves and to people who have the same abilities to be able to determine ones own level of skill. This can be one of the reasons which explains to why cliques are formed.

2.1.3 Heterophily Theory

Without strong ties in an organisation it is unlikely that information and knowledge will flow between the different nodes in a network (Hansen, 1999). Thus it might seem strange that there is a theory regarding the complete opposite, heterophily. As can be read in section 2.1.2 one seeks to balance the relationships either by changing the attitude or by cutting of the relationship as a whole. The heterophily theory suggest that having weak ties to a group of people in a system might be favorable as a sort of ’middle man’ which can interact between two groups of people whom otherwise would be disconnected (Simmel, 1950; Kilduff and Tsai, 2003).

Figure 4: Image which shows two cliques and a heterophilic relationship between them.

2.1.4 Structural Role Theory

Structural cohesion is a part of structural theory which tries to answer ques-tions like "who am I comparing myself with?" or "who am I influenced by?" As read in section 2.1.2 people develop bonds with those who are similar to oneself (they are people are most likely cohesive), and they are structurally cohesive since they are both constrained by the structure of their own group. Their behaviour is limited and changed by other members in their group (Kilduff and Tsai, 2003).

Structural equivalence or role equivalence means that two people may be closely monitoring each other, not because of the structural cohesion, but rather because they play a similar role in the system (or they are in different systems but still have a similar role) (Kilduff and Tsai, 2003).

(13)

2.1.5 Different analytical approaches

There are several ways to analyse social networks. Kilduff and Tsai (2003) describes that one way is to collect whole network data by writing down all the names in the network to be analysed and allowing each and every individual to answer the question "Which one of these individuals do you deem to be close friends with?". That way one could create a matrix over the social networks. If all persons in the networks are not able to answer the question one is able to create an egocentric matrix based on those that could answer the question. Then one could ask individuals about the relationship between his/her friends. Another approach would be to look at records kept by the organisation. From the network different variables can by analysed (e.g. popularity, centrality, importance).

According to Kilduff and Tsai (2003) there are 3 distinct features of social network studies:

1. Network research focus on relations and patterns of the components, not their attributes

2. Research on networks can be done on both micro and macro levels (or sharp/blunt end in an organisation)

3. Network research allows quantitative and qualitative data analysis as well as graphical representations, and integrates these components well with each other.

Other uses of social networks can be to investigate if the friendship (or hierarchy) influence the decision making process in a certain situation. In this study, a matrix is not formed by investigating the friendship between actuators but rather by studying the communication data from a simulation. That way one is able to make a matrix illustrating the importance of a person in an organisation as well as studying the information flow in a certain scenario.

2.2 Social networks and cognition

Kilduff and Tsai (2003) implies the cognition of an individual plays a crucial part when it comes to organisational networks. If a colleague believes that a prominent person in the organisation is your friend, then your colleague will tend to think that you yourself is a higher performer. The perceived friendship is thus a link between your friends status and your own, and

(14)

whats more important is that it is the perception of your relation to your friend and not what the true friendship really is (Kilduff and Krackhardt, 1994). Kilduff and Tsai (2003) believes that this part is quite important in a social network. For example; where does the perceived connection between two individuals start and what are the implications of the perceived con-nections? According to Kilduff and Tsai (2003), most perceived connections starts by observing the behaviour in a networks, and after certain amount of observations, one believes that network relationships should follow a cer-tain pattern, and because of this one could assume that information flows between individuals in an organisation in a certain pattern which is based on the observers earlier perception. Which could be related to (for example) cognitive maps (see Bermúdez (2010); Sternberg (2009)): , but on a social network level. It is extremely important that these maps are accurate since it appears that individuals who are seen as a high status object in a social networks are quite relied on when in comes to making major decisions (Kilduff, 1990), which could (by definition) have devastating consequences.

To summarize one could say that the networks form as a learning process based on perception and observation, whether these networks are mappable to the organisations own framework or planned structure is however, unknown. These cognitive maps over the framework can also differ between different individuals and thus there is a risk for chaotic scenarios when several individuals have to interact with each other.

2.3 Cognitive Networks

Cognitive networks might be seen as an unpromising theory (Kilduff and Tsai, 2003), but there are a certain number of key concepts which are studied in social networks (cognitive balance, cognitive accuracy and cognitive maps). Within the theory of cognitive networks one analyses the actor’s perception of the network, how the perception influence the formation of the network and the mutual influence between networks and cognition.

2.3.1 Cognitive balance

People tend to move towards a cognitive balance. For example: if p has a friendship with q and q are friends with x then p has a tendency to like x as well (for a review, see (Crockett, 1982)), and hence the social network is affected by the cognitive balance. Other studies have shown that people strive towards a balanced relationship since unbalanced relationships can

(15)

lead to instability (Festinger and Hutte, 1954) and nervousness (Sampson and Insko, 1964).

2.3.2 Cognitive accuracy

Cognitive accuracy is about how the perception of a social network is mapped against the reality. For example; a social network might be struc-tured by a management to have a clear hierarchy pattern and information flow, while in reality it might be structured by sentiments within the work group. Through different perception of a social network, many conse-quences might happen in any given situation (Kilduff and Tsai, 2003).

2.3.3 Cognitive maps

A cognitive map (sometimes called a mental model) is an individual’s rep-resentation of, for example, spatial information (Bermúdez, 2010; Sternberg, 2009) or network relations (Kilduff and Tsai, 2003). According to Kilduff and Tsai (2003), one could analyse individuals cognitive maps to study reci-procity and transitivity as well as studying the individual’s own opinion of his or her role in the social network.

2.3.4 Information networks

One important aspect in decision making (especially group decision mak-ing) is influence. A network member who has high influence in a network (i.e. a good position) has the ability to add or withdraw information to influence the decisions which are to be made (Haythornthwaite, 1996). It is thus possible to structure one’s network to be beneficial for the actors and optimize the information flow and usage. However, the amount of information does not matter if none in the network has the ability to put the information to use, and a well-structured network can function as a filter to reduce the chance of information overload (Haythornthwaite, 1996). A network has thus two key features; members who can access information and members who can receive and forward the information to other actors.

2.4 Social network analysis

Social network analysis focuses on the relationship and resource sharing between actors in a given system. The resources can literally be anything

(16)

that can be shared between actors (e.g. money, goods, information, influ-ence) (Haythornthwaite, 1996). The actors can be seen as individuals or entire organisations (Haythornthwaite, 1996). In an analysis, the social networks are illustrated as a graph where the relationship or information flow between individuals or other entities are represented by connections (edges) between actors (nodes) and the analysis itself is based on techniques drawn from the mathematical field and from graph theory (Houghton et al., 2006). Five principles that are often analysed are (Haythornthwaite, 1996):

• cohesion (grouping of actors depending of relationship) • structural equivalence (grouping of actors depending of role) • prominence (analyze to find the "person in charge" (actor in charge)) • range (range of an actors network)

• brokerage (showing connection (actors) which bridges to other net-works)

As described earlier, social network analysis focuses on the relationship between actors and the patterns in interaction. This means that social net-work analysis can be used to study almost any phenomena where there is an interaction (or lack there of) between actuators in a system. In other words, social network analysis should be a valid method for the study of information exchange (Haythornthwaite, 1996). Haythornthwaite (1996) explicitly number five key features which can be studied with a social network analysis:

1. Information Needs - study what kind of information a certain actor in a network needs by studying current information change, can be used to distinguish groups or actors, can be used to find structural equivalence or holes in the network.

2. Information Exposure - study information exposure, might be able to indicate prominence and range.

3. Information Legitimation - measure the degree of an actors ability to legitimate the information which they share. Being able to legitimate ones information increases the likelihood of being able to effect the networks decision making.

4. Information Routes - studies the flow of information and which routes it takes to reach the target. Information becomes useful to an ac-tor by being forwarded to others. The patterns which arise from

(17)

from communication can be studied in a social network analysis and thus explicitly showing the already know connections as well as the creation of new information networks.

5. Information Opportunities - That brokerage exists shows that actors are able to take control over information flow in a network by changing their position.

2.4.1 Mathematical operations

One of the more common methods if to first collect data about the relation-ship between actors in the system which is to be studied. The relationrelation-ship can be anything from individuals representation of their relationship to each other to communication. The information gathered is then placed in a matrix. An example matrix can be found in table 1 below.

Table 1: An example of a relationship matrix which can be used for a graphical representation or mathematical operations

Steve Clint Bruce Clark Natasha Barry Wade

Steve 0 8 7 0 12 0 5 Clint 6 0 3 0 4 0 9 Bruce 6 2 0 0 0 13 6 Clark 1 0 0 0 0 14 4 Natasha 3 4 6 0 0 0 7 Barry 1 0 0 13 0 0 5 Wade 3 4 2 4 7 3 0

The matrix can after that be used to conduct mathematical operations or used to construct a social network graph (Houghton et al., 2006). The mathematical operations have certain benefits over the visual illustrations (like inhibiting the intuitive reading of the graph or a form of data mining when the data network data is large or complex) (Houghton et al., 2006). The mathematical operation should be chosen dependent on the data and the aim of the analysis (Wasserman and Faust, 1994).

A mathematical operation which can be applied to social networks are the sociometric status, which measures how much ’activity’ a node has in relation to all other nodes. The higher the score, the more active a node is in a network. The equation for sociometric status is the same as used by Houghton et al. (2006):

(18)

sociometric status = 1 g−1 g

∑

j=1 (xij+xji)

In the above equation, g represents the total number of nodes in the network, i and j are individual nodes and xijrepresent the edge values from

node i to node j and vice versa for xji.

To further explore a network through mathematical methods, a mea-surement of centrality can also be used to calculate the standing of a node. Centrality basically shows if a node is close to other nodes or not (it reflects the degree of access to information a node has). Although similar to so-ciometric status they are not the same since a node with high soso-ciometric status can be found in the peripheral part of a network (Houghton et al., 2006), and thus both sociometric status and centrality should be calculated since there is a chance that a node with high sociometric status does not have a high centrality metric, another benefit of the centrality measurement is that it can be used at both group and individual level (Pfautz and Pfautz, 2009). The formula below is an example algorithm (Bavelas-Leavitt) to calculate centrality, also previously used by Houghton et al. (2006):

Bavelas−Leavitt centrality= ∑

g

i=1;j=iδij

∑g_j=1(δij+δji)

The g in the equation above symbolizes the size of the network, δjiis the

geodesic distance between the nodes.

Another centrality measurement which can be used as a complement to Bavelas-Leavitt is the betweenness centrality which is a measurement that shows how much interaction one node has with all other members in the network (Schraagen and Post, 2014). The following is the equation for betweenness centrality: Betweenness(i) = g

∑

j=1;k=i GPathsj→i→k Gpathsj→k

2.4.2 Classic network structures

Earlier studied in the field of social networks have shown small number of network structures. These network are labeled as 1) circle, 2) chain, 3) Y, and 4) wheel networks. These earlier studies have, for example, shown that a wheel network works best for simple problem-solving tasks while

(19)

the circle network resulted in worst performance (Leavitt, 1951), although this study shows that the wheel network had quite a high error rate, this is believed to be caused by a human misunderstanding of the task in the network and not due to the network itself. One should note that the wheel structure only works for simpler tasks, as stated by Houghton et al. (2006), the middle person in the wheel might be overwhelmed by the information load which is heading his/her way due the the position in the middle of the network, and as such the circle structure might be preferable (although there are some cases to be discussed about the implications of the circle structure, such as information loss). Figure 5 show the strategies found by Leavitt (1951).

Figure 5: Figure shows the classic strategies found by Leavitt (1951). A study have shown that a naval team in training adopts a sort of wheel network where each side of the wheel are connected (Schraagen and Post, 2014) (see Figure 6, which might reduces the information load of the center piece while at the same time increase the system resilience. In other words; by studying network structures one can answer different question then by using mathematical equations.

Figure 6: Figure shows a strategy found by (Schraagen and Post, 2014) after a naval team been practicing together.

(20)

2.4.3 Other network approaches

There are other network approaches which might be similar to social net-works but should be distinguished from recent structural approaches and shared team knowledge (Schraagen and Post, 2014) which focus on surveys to measure team knowledge (Espinosa and Clark, 2013; Avnet and Weigel, 2012) and not real time communications. Social Network Analysis has also nestled its way into the field of ergonomics (in methodologies such as EAST (Walker et al., 2010)) where the authors tries to show how EAST method is connected to the perspective of distributed cognition.

(21)

3 Method

3.1 Simulation

The datasource used in this thesis if from a simulation of a major incident. The main goal of the simulation was to; 1) Train command functions, 2) Train rescue services when dangerous goods (i.e. toxic, flammable material) are involved, 3) Make sure that the emergency plans (instructions) which currently exist are correct, and 5) That the results from the simulation can be used further by each organisation. The simulation were of an accident classified as an CBRNE-accident (CBRNE - Chemical, Biological, Radiological, Nuclear, Explosive).

3.1.1 Scenario

The simulation scenario, which took place in Norrköping, took place at a level crossing over Blixtholmsvägen. In the scenario a petrol truck, a minivan and a train had a collision, several were injured and Acrylonitrile (a toxic, flammable chemical liquid substance which vapors and direct contact are damaging to the human body) began leaking from one of the train cars. The substance (and the substance vapors) in the petrol truck are also toxic and highly flammable.

The full scenario was as follows (Leonardsson and Johansson, 2015): A train (model T44), which pushes tanks with hazardous substances infront of it, hits a heavy vehicle which is at the side of the train track. Further along the road the collides with a minivan (containing students on a field trip) and a petrol truck (which also contains hazardous substance) at a level crossing and the minivan gets stuck between the train and petrol truck (at this point, a train cart containing Acrylonitrile starts to leak). In the current scenario, maintenance was currently being conducted on the train track and thus no warning signals were sent to the minivan nor to the petrol truck which were at the level crossing. The maintenance workers did notice the train and moved away from the train track, but for some reason the heavy vehicle which were used in the maintenance work was not far enough away and got hit by the train. Behind the minivan was a regular school bus carrying several students on the way to the same field trip as the minivan. Figure 7 illustrates the accident layout of the accident.

(22)

3.1.1.1 Sequence of events

The following item list describes key events in the simulation in chronolog-ical order. The following list is not a complete description of the scenario and should only be used to get an understanding of the development of the situation.

• 10:18 - Call from the PSAP (emergency dispatch center) regarding the accident.

• 10:22 - Police incident commander, incident commander and ambu-lance incident commander confirms their presence to each other. • 10:24 - Rescue service arrives at the scene.

• 10:32 - Ambulance incident commander and Medical incident com-mander arrives at the scene in the first ambulance.

• 10:41 - First meeting between the joint incident command (Incident commander, police incident commander and ambulance incident commander).

• 10:52 - Second ambulance arrives at the scene.

• 10:52 - Medical incident commander updates Ambulance incident commander regarding the medical condition of those who were in-volved in the accident.

• 10:56 - Second meeting of the joint incident command. • 11:03 - First patient get transported to the hospital.

• 11:07 - Medical incident commander appoint the personnel from the second ambulance as sector leaders for the assembly point.

• 11:15 - Second patient is transported from the scene. • 11:21 - Third meeting by the operative leaders. • 11:23 - Third patient is transported from the scene. • 11:52 - Fourth meeting by the operative leaders. • 12:27 - Rescue service initiate decontamination.

(23)

(24)

3.1.2 Participating organisations and goals

Following organisations participated in the simulation ; 1) Rescue ser-vices, 2) Police, 3) County council Östergötland (LiO, including ambulance services), 4) PSAP, 5) Towing employees, 6) Railway company, 7) Traffic control (Veolia Transport AB), 8) Railway maintenance, 9) Styron Sverige AB, 10) Municipality of Norrköping, and 11) County administrative board (Länsstyrelsen). However, not all of the above were of importance in the main simulation. The focus of the simulation were to get the different organisations in the society to cooperate in a rescue operation as well as improving the cooperation between cilvil protection, healthcare, police and contractors thus the data collection focused on the viewpoints from (al-though, all interactions which were recorded in the simulation are included in the data analysis):

• Municipality of Norrköping • Rescue Service • Traffic control • Police • LiO

Municipality of Norrköping

The goal of the municipality of Norrköpng was to have a overarching and coordinating role. Other than that, the municipality of Norrköping prac-ticed the psychosocial care which is expected by them thorough political decision. The municipality of Norrköping should also collaborate with the police and county council crisis group as well as the county councils medical personnel.

Rescue service

The main goal for the rescue service was to train the readiness to handle larger accidents involving dangerous goods. Besides that the rescue service wanted to investigate the function of the emergency pathways, that the cooperation between different organisations worked as intended and that the command levels works in a larger accident. Above that the rescue service wanted to investigate the interoperability between the different staff organisations when it comes to establishing the accident zone.

(25)

Traffic control

The goal of the traffic control was to make sure that the emergency path-ways works as intended as well as making sure that the commanding officer (from this organisation) who arrives at the scene follow the expected routines. The goal of the traffic control headquarters is to make sure that the emergency pathways works and that support persons are selected and that contact is established with the relatives of the wounded.

Police

The main goal of the police organisation was to practice the organisation when it comes to handle an accident of a larger scale. The different authori-ties and organisations are trained in conjunction at the scene of the accident as well as in other locations (e.g. communication central).

LiO

The main goal of LiO was to improve the county councils ability to handle an accident where dangerous substances are involved (a so called CBRN-accident). Their goals included to evaluate pre-hospital medical care on the scene of the accident, as well as evaluating that the person in charge at the ER follow the regulation at the hospital.

3.1.3 Data collection

Data was collected from various sources, and not by the author of this thesis. Audio data were collected through both a lavalier microphone as well as through RAKEL (a national TETRA-based communication system used by emergency services and other fields of rescue service, public safety and security, emergency medical services and healthcare). The audio was recorded from two different organisation roles; 1) Ambulance incident commander, and 2) Medical incident commander.

The video data was collected by using cameras mounted on those who had a "leadership responsibility" at the incident scene. The cameras were recording from these views; 1) Rescue service, 2) Ambulance incident commander, 3) Medical incident commander, and 4) Incident commander. However, to ensure that the social networks which were to be created was only based on high quality data, the networks are based on the Ambulance incident commander as well as the Medical incident commander since

(26)

the audio uptake was significantly worse from the incident commander’s viewpoint (since the incident commander was not equipped with an audio recording device), the incident commander was excluded from the study.

3.1.4 Missing data

The dataset should be regarded as incomplete. There is a difference in the amount of data between the ambulance incident commander and the medi-cal incident commander (the data from the medimedi-cal incident commander was cut short both audio and video). The video segments were able to complement some of the missing audio, but it was still not enough to be on the same level as the ambulance incident commanders. Because of this, the last section of the ambulance incident commanders data has been removed from the analysis.

3.1.5 Data analysis

The data analysis was divided into 4 parts: 1. Transcription and filtering of data.

2. Network creation (Six networks were created in total; 1) Ambulance incident commander, 2) Medical incident commander, and 3) a con-joined network between the two for both filtered and unfiltered data). 3. Network calculation.

4. Comparing real networks with expected network.

The networks presented in this thesis were created using Agna software. The software are able to create a graphical network as well as calculating the sociometric status and centrality for each node in the network.

3.2 Procedure

The data used in this study were already collected and consisted of audio and video material from a medical catastrophe simulation in the Munic-ipality of Norrköping. The first step was to transcribe the data based on interactions (i.e. when one individual interacted with another). The transcription included a time stamp (when the individuals interacted), in-teraction initiator and receiver, as well as a description of the inin-teraction (i.e. who said/did what). For example, a data cell could look like this:

(27)

0:53:21 Ambulance incident commander -> Medical incident commander | "For your knowledge, one ambulance has arrived" (Information sharing, no explicit demand from Medical incident commander)

Since it can be quite hard to identify individuals from voice alone, the audio recordings were complemented by video recordings (although they had to be synchronized by hand). In total 2 hours and 33 minutes were transcribed from the Ambulance incident commander’s audio recordings and 1 hours and 47 minutes were transcribed from the Medical incident commander. In other words there is a difference in the amount of audio data which were recorded and the final segments of the audio recordings were complemented by video recordings (although the quality of the audio were lower). It should be noted that even if the audio recordings of the Medical incident commander were complemented by video, the amount of data still did not synchronizes with the Ambulance incident commander since the last segments were still missing. The network which were to be created is then, by definition, classified as a partly incomplete network, and the focus of the analysis will be on the medical services. But since the data from both study objects are nearly 2 hours long, the dataloss should be considered as negligible. To make sure that the network would be valid, the data from the Ambulance incident commander was cut from the point where the data from the Medical incident commander ended. Due to the amount of interaction with the rescue service as well as the police force, the interaction with the police and rescue service was partly analysed, but only from the view point of the Ambulance incident commander as well as the Medical incident commander.

After the material were transcribed the data were coded into two higher levels of categories, information request and information share. After the higher levels of categorising the following subcategories were found:

• Accident information (information regarding the accident). • Safety (information regarding the safety at the accident site). • Current status (current validated information and current plan). • Future actions (actions which are planed).

• Commanding (when a command is distributed).

• Organisation (information specific to one organisation (e.g. number of wounded or triage information)).

(28)

• Confirmation (confirming an action).

• Resources (information regarding various resources (e.g. number of ambulances)).

• Organisational structure (information regarding the structure of the organisation (e.g. who to contact)).

• Aid (requesting aid from another organisation, but not classified as a command).

• Validating (validating information).

• Health (information concerning the current medical status of an indi-vidual).

• No information (when information request was denied or no infor-mation could be given).

By using the above filter criteria a new network graph and relationship matrices representing the communication pattern for both the Ambulance incident commander and Medical incident commander were created. This filter was used to make sure that no noisy data were included in the analysis (e.g. if one person merely greets another)

Graphs were created based on interaction matrices (earlier called rela-tionship matrix) based on real time communication recorded during the simulation. If person X spoke to person Y (node X and node Y in a graph), that vill increase the edge value between X and Y by 1 (i.e. increase the value inte the matrix by one). If one person spoke to a group of people, it was counted as one interaction from that person to each group member. The analysis was done on unfiltered data (including all interaction between the nodes in the network) and filtered data (only including relevant interaction between the nodes).

All graphs were created in the software Agna (version 2.1). After the two networks had been created and analysed separately, the networks were conjoined with each other to be able to investigate the organisational structure by analysing the two different roles in relation to each other. Since the networks were build in the software Agna, calculating the sociomet-ric status and the two different centrality measurement (see section 2.4.1) for each node was done by the program and thus minimizing the risk of human error in the mathematical calculation. The graphs and the mathe-matical results was then compared to the expected structure derived the the organisations own rules and regulation document.

(29)

4 Result

The following section contains the result from both the social network anal-ysis based on real-time communication as well as the expected interaction pattern based on protocols from the organisations.

4.1 Social Network analysis - unfiltered

The following sections will present a graphical network and network data based on unfiltered transcription data. This means that all "irrelevant" information will be included in these sections (i.e. If the medical incident commander said "Hello" to another person, this counts as 1 interaction and will add 1 to the edge value).

4.1.1 Ambulance incident commander

From the relationship matrix based on the ambulance incident comman-der’s interaction, the following graph were created.

(30)

Figure 8: Visual graph representing interaction between nodes from the ambulance incident commander’s point of view

(31)

From the data gathered from the simulation the following mathematical data was extracted:

Table 2: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality

Node (actor) Sociometric status Bavelas-Leavitt centrality Betweenness centrality

PSAP 1.6428572 7.018182 0.0

Medical incident commader 25.857143 13.310345 163.0

Medical incident commander 9.642858 7.018182 0.0

Incident commander 1 9.785714 8.391304 7.5

Designated duty officer 3.7142856 7.018182 0.0

Police incident commander 6.428571 7.877551 0.0

Division commander 0.42857143 7.018182 0.0 Care provider 9270 2.0714285 7.018182 0.0 Police officer 0.14285715 7.018182 0.0 Care provider 9720 0.21428572 7.018182 0.0 Care provider 9260 0.35714287 7.018182 0.0 Care provider 9530 0.14285715 7.018182 0.0 Incident commander 2 1.7857143 8.041667 0.0 Traffic control 1.8571428 8.212766 5.5 Green cargo 1.0714285 6.6551723 0.0

As the table above shows, from the ambulance incident commander’s point of view, a large amount of interactions between other’s in the net-work takes place during the simulation (sociometric status = 25.857143). The ambulance incident commander also had a higher Bavelas-Leavitt centrality (13.310345) and i higher betweenness centrality (163.0) than the rest of the actors which means that the ambulance incident commander has great opportunities to gather information as well as interacting with many different nodes in the network. Although lower values, the other cooperative leaders in the network also have relative high values in the three categories.

4.1.2 Medical incident commander

From the relationship matrix, the following graph was created to represent the interaction between the medical incident commander and the nodes which the commander had interacted with during the simulation.

(32)

Figure 9: Graph representing interaction from the medical incident com-mander’s point of view

(33)

Using the Agna software, the following data was extracted from the relationship matrix:

Table 3: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality, based on the data from the medical incident commander

Node (actor) Sociometric status Bavelas-Leavitt centrality Betweenness centrality Medical incident commander 14.121212 30.84127 920.0

Other personnel 0.030303031 32.383335 0.0

Ambulance incident commander 4.121212 16.057852 32.0

Division commander 2.121212 16.191668 0.5

Rescue service 1 0.6363636 15.796748 0.0

Rescue service 5 0.060606062 15.796748 0.0

Green cargo (wounded) 0.15151516 15.796748 0.0

Civilian (wounded) 1 0.060606062 15.796748 0.0 Civilian (wounded) 2 0.060606062 15.796748 0.0 Civilian (wounded) 3 0.060606062 15.796748 0.0 Civilian (wounded) 4 0.060606062 15.796748 0.0 Civilian (wounded) 5 0.060606062 15.796748 0.0 Civilian (wounded) 6 0.060606062 15.796748 0.0 Anker AB 1 0.27272728 15.796748 0.0

Anker AB (severely wounded) 2 0.060606062 15.796748 0.0 Rescue service (healthcare) 0.8181818 15.9262295 0.0

Gravely wounded 0.030303031 32.383335 0.0 Anker AB supervisor 0.121212125 12.699347 0.0 Care provider 9270 0.3030303 15.796748 0.0 Care Provider 9720 1 0.969697 15.796748 0.0 Care provider 9720 2 1.4545455 16.057852 0.0 Care provider 9260 0.030303031 32.383335 0.0 Care provider 9530 0.3939394 15.796748 0.0 Rescue service 3 0.54545456 15.796748 0.0 Rescue service 4 0.54545456 15.796748 0.0 Police officer 4 0.121212125 16.057852 0.0 Police officer 1 0.3030303 15.796748 0.0 Police officer 2 0.060606062 15.796748 0.0 Police officer 3 0.060606062 15.796748 0.0 Care provider 9190 0.36363637 15.796748 0.0 Police officer 5 0.15151516 15.796748 0.0 Traffic Control 0.060606062 15.796748 0.0

As with the ambulance incident commander, the medical incident com-mander handled a significant amount of interactions (sociomentric status = 14.121212, had a position which allowed greater amount of information (Bavelas-Leavitt = 30.84127) and also interacted with a high amount of

(34)

nodes (Betweenness centrality = 920.0). The actor labeled as "Other per-sonnel" is only interacted with at one point in the simulation (when the medical incident commander is leaving the garage) but the formula used to calculate how much access to information one has still labels this node as highest, which is a topic for disussion.

(35)

4.1.3 Joint interaction

If the relationship matrices from the ambulance incident commander and medical incident commander merged into one single matrix, the network would expand. The following figure represents the merged network.

Figure 10: Graph representing the medical incident commanders and am-bulance incident commanders join network

From the relationship matrix from the joined data, the following mathe-matical data was extracted from the Agna software and presented in Table 4.

(36)

Table 4: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality, based on data from both the medical incident commander and ambulance incident commander.

Other personnel 0.02631579 37.135803 0.0

Rescue service 1 0.55263156 18.91824 0.0

Rescue service 5 0.05263158 18.91824 0.0

Gravely wounded 0.02631579 37.135803 0.0 Anker AB supervisor 0.10526316 16.99435 0.0 Care provider 9270 1.0263158 20.744827 0.0 Care Provider 9720 1 0.84210527 18.91824 0.0 Care provider 9720 2 1.3421053 21.034966 6.0833335 Care provider 9260 0.15789473 17.694118 0.0 Care provider 9530 0.39473686 20.744827 0.0 Rescue service 3 0.47368422 18.91824 0.0 Rescue service 4 0.47368422 18.91824 0.0 Police officer 4 0.10526316 19.159235 0.0 Police officer 1 0.2631579 18.91824 0.0 Police officer 2 0.05263158 18.91824 0.0 Police officer 3 0.05263158 18.91824 0.0 Care provider 9190 0.31578946 18.91824 0.0 Police officer 5 0.13157895 18.91824 0.0 Traffic Control 0.7105263 21.956203 39.483334 PSAP 0.6052632 15.666667 0.0

Green cargo (leader) 0.39473686 15.505155 0.0

Police officer 6 0.05263158 15.666667 0.0

From the table above one can derive that that Medical incident comman-der had by far the largest sociometric status (12.263158) and also the highest

(37)

betweenness centrality (1141.1167). The second highest was the ambulance incident commander with a sociometric status around 9.5 and a between-ness centrality around 352. The one who, by the software, were deemed to have access to the most information was the "Gravely wounded" and the "Other personnel" node. When joining the two networks and studies the medical staff, one notices that the medical incident commander (when merging the two networks) has higher values in all the different measure-ment. This indicates that this actor has access to a lot of information in the network, even more so than the ambulance incident commander. The difference in information structure is, however, still unclear.

4.2 Social Network analysis - based on filtered data

The results in the above subsection is solely based on unfiltered data. This means that every kind of interaction, relevant or not, in the simulation are represented in the data. To broaden the analysis and to further explore the validity of the method, the results in this subsection is based on filtered data. This means that all the noise which might exist in the previous section have been removed and only information sharing and information receiving is included.

4.2.1 Ambulance incident commander

If one removes all the noise from the data and only include interactions which are based on information which is relevant, the network looks the same.

(38)

Figure 11: Graph representing the network from the ambulance incident commander’s point of view. Based on filtered information

And as one can see from the table below, there are several changes in the mathematical data which can be derived from Agna.

(39)

Table 5: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality from the ambulance incident commanders filtered data

Node (actor) Sociometric status Bavelas-Leavitt centrality Betweenness centrality

PSAP 1.2142857 6.981818 0.0

Medical incident commander 7.428571 7.111111 0.0

Division commander 0.64285713 7.111111 0.0 Care provider 9270 1.1428572 6.981818 0.0 Police officer 1 0.14285715 6.981818 0.0 Care provider 9720 0.21428572 6.981818 0.0 Care provider 9260 0.5714286 6.981818 0.0 Care provider 9530 0.14285715 6.981818 0.0 Incident commander 2 1.8571428 8.170213 0.33333334 Traffic control 1.8571428 8.170213 5.5 Green cargo 1.0714285 6.62069 0.0

The results now show lesser amount of information in the network, but the ambulance incident commander still ranks highest in all three categories. Although the care providers seem to have access to a lot of information, there are no values indicating a higher amount of communication passing to these nodes (which is the same as in the previous unfiltered networks).

4.2.2 Medical incident commander

The following network represents interaction between the nodes without any sort of noise in the information sharing.

(40)

Figure 12: Graph that represents the medical incident commanders social network without any noise in the data

The table below shows the sociometric status, bavelas-leavitt and be-tweenness centrality for each node.

(41)

Table 6: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality from the medical incident commanders filtered data

Rescue service 1 0.5483871 14.939655 0.0

Rescue service 5 0.032258064 30.40351 0.0

Anker AB supervisor 0.12903225 14.939655 0.0 Care provider 9270 0.22580644 14.939655 0.0 Care Provider 9720 1 0.87096775 14.939655 0.0 Care provider 9720 2 1.0967742 15.201755 0.0 Care provider 9260 0.032258064 30.40351 0.0 Care provider 9530 0.32258064 14.939655 0.0 Rescue service 3 0.2580645 14.939655 0.0 Rescue service 4 0.2580645 14.939655 0.0 Police officer 4 0.12903225 15.201755 0.0 Police officer 1 0.2580645 14.939655 0.0 Police officer 2 0.16129032 14.939655 0.0 Police officer 3 0.06451613 14.939655 0.0 Care provider 9190 0.2580645 14.939655 0.0 Police officer 5 0.16129032 14.939655 0.0 Traffic Control 0.06451613 14.939655 0.0

When looking at the filtered medical incident commander’s network one can still see that the Medical incident commander is superior when it comes to sociometric status and betweenness centrality. But is loses to one care provider and one node from the rescue service when it comes to the bavelas-leavitt centrality even though they are almost never interacting.

(42)

4.2.3 Joint interaction

The following graph represents the the network which are based on both matrices from the ambulance incident commander and the medical incident commander.

Figure 13: Graph that represents the joint network between the medical incident commander and ambulance incident commander, based on filtered data

The following table contains the mathematical data from the relationship matrix.

(43)

Table 7: Data of sociometric status, Bavelas-Leavitt centrality and Between-ness centrality from the joint network, based on filtered data

Node (actor) Sociometric status Bavelas-Leavitt centrality Betweenness centrality Medical incident commander 10.485714 32.692307 1020.48334 Ambulance incident commander 8.028571 22.368422 205.13333

Rescue service 1 0.4857143 17.586206 0.0

Rescue service 5 0.028571429 34.45946 0.0

Anker AB supervisor 0.114285715 17.586206 0.0 Care provider 9270 0.6 18.88889 0.0 Care Provider 9720 1 0.7714286 17.586206 0.0 Care provider 9720 2 1.0571429 19.172932 4.116667 Care provider 9260 0.25714287 16.13924 0.0 Care provider 9530 0.34285715 18.88889 0.0 Rescue service 3 0.22857143 17.586206 0.0 Rescue service 4 0.22857143 17.586206 0.0 Police officer 4 0.114285715 17.832169 0.0 Police officer 1 0.22857143 17.586206 0.0 Police officer 2 0.14285715 17.586206 0.0 Police officer 3 0.057142857 17.586206 0.0 Care provider 9190 0.22857143 17.586206 0.0 Police officer 5 0.14285715 17.586206 0.0 Traffic Control 0.7714286 19.921875 28.15 PSAP 0.4857143 15.9375 0.0

Green cargo (leader) 0.37142858 14.010989 0.0

The table of the joint networks between the ambulance incident com-mander and the medical incident comcom-mander reveals several key features of their interaction. In the joint network the medical incident commander is above all others in the three different measurements when it comes to isolated information sharing and receiving. The ambulance incident com-mander is at the second ranking which shows that from their view, these

(44)

two individuals are of most importance in the simulation. The different care providers have access many information opportunities, but appears to not be taking advantage of these opportunities. The first incident commander and the police commander appears to also have an important role com-pared to the rest of the accident personnel, but they cannot be comcom-pared without proper data.

4.3 Organisational rules and regulations - results based on

protocols

4.3.1 Rescue service

In the organisation of rescue service there were always be an incident commander at the scene of a CBRNE-accident. The incident commander is, at a municipal level, also the rescue manager, and can thus appoint the role of incident commander to someone else. If the rescue manager choses to appoint someone other than himself as the incident commander the rescue manager will still have the utmost responsibility at the scene of the catastrophe and the newly appointed incident commander will have to abide the rules and command of the rescue manager.

The incident commander is often the one in charge of a rescue operation, and will more often than not be at the scene of the accident. The incident commander will initiate and end the rescue operation and can make deci-sion which inflicts on another persons right (for example: block or evacuate an area). There is not standardized organisational structure, but the most common structure is illustrated in figure 14.

(45)

Figure 14: A graphical representation over the rescue service organisation In the structure above another role has emerged at the scene of the accident. In this organisation the incident commander will spend most of the time at the leading site while the incident site manager will command the rescue operation at the scene of the accident and will command the different sector managers (a sector is an arbitrary term which can be based on geographical locations or different teams which have been constructed on the scene).

The rescue service are responsible for the security in case of an CBRNE-accident. This means that the rescue service will designate an arrival route (the area where the different units from different organisations will stand on standby in waiting for assignments) roadblocks and zone classification. The rescue service will support the medical care units as well as the police force by giving information about safety distance or protective measures against the harmfull substance which might be involved in the accident. The rescue service is also the organisation who orders which protective gear that needs to be used at the accident scene.

(46)

questions below different decision regarding the safety will be made. • What substance is involved?

• How much substance?

• Is the substance outside its container?

• Are there any people at, or around, the scene of the accident? • What is the surrounding environment?

• What is the current and expected wind heading?

4.3.2 Police

The police is regulated by law to cooperate with other organisations in case of emergency. A couple of ways of doing so is to create road blocks or limit the access to certain areas and other operations which aim to aid the incident commander. The police share the responsibilities of security with the rescue service organisation. In case of CBRNE-incidents there are four main items which they are responsible for:

1. Save lives.

2. Protect the health of citizens and properties/environment. 3. Preserve evidence.

4. Restore the scene to normal.

In case of special incidents (like a CBRNE-accident) the police manage-ment consist of:

• The strategic management.

• The general operational management. • The tactical management.

The tactical management (also called police incident commander or op-eration commander) is the only one of the three above which is at the scene of the accident and is determined by the one who is classified as a police commander. The field staff is determined by the operation commander. Figure 15 shows a graphical representation over the police organisation.

(47)

Figure 15: A graphical representation over the police organisation The police incident commander (who is at the accident scene) have the following responsibilities:

• Estimate risk conditions. • Chose a suitable tactic.

• Apply situation-based conflict management and chose between dif-ferent tactical approaches.

(48)

• Be able to participate in the decision making from the strategic man-agement.

• Chose a suitable management model.

• Collect information which is needed for the assignment and cross-check this information with the shared situational awareness.

• Design the assignment.

• Make sure that the task force at the scene have an adequate amount of stamina.

• Judge the competence and suitability of the personnel at the scene. • Cooperate with the other organisations at the scene (medical

person-nel/ rescue service etc.).

• Mental preparation, stress management and crisis management. • Document important decision.

• Appoint field staff.

4.3.3 Healthcare

The healthcare system is designed to handle "normal" situations. In the case of an extraordinaire accident the structure of the system might need to be redesigned. In those events the healthcare is under the command of the designated duty officer (TIB).The healthcare team leads and are responsible for the health related work at the scene of an accident but can be limited in moment by other organisations (such as the police or rescue service).

In the field, medical personel can have one of four possible roles (the first three are mandatory):

1. Ambulance incident commander. 2. Medical incident commander. 3. Care providers.

4. Sector leader.

(49)

Figure 16: A graphical representation over the healthcare organisation The designated duty officer is responsible for delegation of the distri-bution key (order of where to transport the different patients based on their wounds) and can answer different medical questions from the ac-cident scene. The Ambulance inac-cident commander cooperates with the incident commander and the operation commander and is the only one in the medical team to communicate with the higher level in the organisations. The Medical incident commander makes sure that the healthcare given follows the rules which are applied in the area of the accident and com-municates with the Ambulance incident commander regarding resources

Using social network analysis as a tool to create and compare mental models