Temporal changes of team networks
during collaborative task performance
Rebeka O. Szabo 11146362
o.sz.rebu@gmail.com
Sociology, Msc: Comparative Organizations and Labour Studies
1st supervisor: dr. Jeroen Bruggeman 2nd supervisor: dr. István Síklaki
Master Thesis
Sociology - Comparative Organizations and Labour Studies Graduate School of Social Sciences
TABLE OF CONTENT
Abstract ... 1
Acknowledgements ... 1
1. Introduction ... 2
2. Theoretical context ... 5
2.1. Network structure and information flow ... 6
2.2. Pulsing interactional pattern and exploration spaces ... 7
3. Research design ... 12
3.1. Research field and method ... 12
3.2. Approach and data preparation for the analysis ... 13
3.2.1. Interactions as network ties ... 14
3.2.2. Dependent variable ... 15
3.2.3. Research sample ... 16
3.2.4. Additional data collecting tool and further analysis plans ... 16
4. Analysis and results ... 19
4.1. Temporal network changes ... 19
4.2. Interactional intensity ... 36
4.3. Other variables ... 37
5. Discussion and conclusion... 59
Bibliography ... 63
Annexes ... 65
Questionnaire sample ... 65
1 ABSTRACT
Teamwork is based on collaboration that permits to achieve higher and more complex goals that would transcend individual capabilities. Well-functioning teams can be considered as basic pillars of successful organizations. Thus, collaboration in team performance is a topic of interest in numerous organizational studies and areas of social and behavioural sciences. Nevertheless, the greatest number of these scientific inquiries follow a research design relying on the investigation of static information gained by the examination of input-output factors of cooperating teams, by which treatment they tend to overlook actual interactional process of collaborative task-solving, and merely able to make inferences about what is going on in between the manipulated input-output variables. The thesis intends to bring a novel approach to this discourse by implementing temporal analysis of collaborative teams’ interactional pattern in terms of successful performance, through the investigation of how teams use their social networks during the time of project-accomplishment. I conducted the data collection by video recording of ongoing interactions of naturally occurring groups in the
semi-laboratory experimental circumstances of an escape room that environment allows to observe small project teams’ collaborative activity. Using various methods and techniques of R statistical programming environment during the analysis of the 16 examined groups (with fifty-fifty ratio regarding successful performance), the first hypothesis claiming that
homogenous distributional dense temporal network structure (being formed by interactions among members) fosters problem-solving is confirmed. While the second hypothesis
suggesting a pulsing interactional pattern to be positively influential on task performance has not gained support by the available data. Beside the central findings, teams’ compositional and relational network characteristics were also explored in correlation with successful project-accomplishment.
ACKNOWLEDGEMENT
First, I would like to thank my thesis supervisor dr. Jeroen Bruggeman of the Faculty of Social and Behavioural Sciences at the University of Amsterdam. He consistently allowed this paper to be my own work, but steered me in the right direction whenever I ran into a
problematic spot. I would also like to acknowledge dr. Istvan Siklaki of the Faculty of Social Sciences and International Relations at Corvinus University as the second reader of this thesis, and I am grateful to him for his helpful suggestions. I would also like to thank the priceless expert contribution to the statistical analysis to Janos Gutmayer. Without his passionate participation and input, such a complex analysis could not have been successfully conducted. Last, but not least, I have to express my gratitude to Attila Giday, technical and operating manager of Paniq room, the research field, for providing me with fundamental conditions being indispensable to data collection and to the initial observations. This accomplishment would not have been possible without him.
2 1. INTRODUCTION
Collaboration between individuals allows groups to combine personal task-solving capabilities, thereby attain higher, cooperation-demanding goals that would be hardly achievable without synthesis of co-workers’ personal knowledge, skills and capacities. Collaboration is a primal, inherent feature of group work activity by establishing mutual interactions and interdependences between members, besides merges their personal capabilities to a greater, joined resource. A group is determined as “two or more individuals who are connected by and within social relations,” while collaboration with a goal to be achieved accounts for an “organized, task-focused group,” defined as a team. (Forsyth, 2009:3&352).
The level of collaboration influences quality of performance, thus the outcome of teamwork, while cooperating activity itself involves two main contributing factors: the compositional and structural settings of collaborative groups. Compositional factors are specified by several variables such as age of the members, the size of the group ergo group characteristics that are primarily originated in individual features of members that influence and form the setup of the team. Structural properties are the “underlying pattern of roles, norms, and networks of relations among members that define and organize the group.” (Forsyth, 2009:144.)
Collaborative team processes with regard of the interplay between input factors, that are compositional and structural settings of teams and group performance that is also called as the output variable have been examined by numerous, managerial and organizational studies just like being the topic of interest of many fields of social and behavioural sciences. However, most of these researches either measure interactional process by retrospective questionnaires, or rely on an input-output design, thus focus on the two endpoints of group activity, thereby tend to overlook tremendous meaningful data on actual ongoing processes and links that they originally meant to observe.
Retrospective questionnaires collect group members’ ratings or reports about what happened in the team previously. As a consequence, information yielded by retrospective questionnaires is necessarily coarse-grained data, thus, it gives entirely subjective information which is likely to include biases in connection with heuristics such as highlighting the earliest or most salient interaction at the expense of all the interactions. Moreover, these reports of team members measured by retrospective questionnaires are likely to be also influenced by prior group life, than merely summarize what went on a certain group occasion objectively. Consequently, retrospective questionnaire is an inappropriate choice to collect fine-grained interactional
3 data; they are not even designed to do so, however, this is a commonly applied method to collect information of group processes.
Input-output research method, on the other hand is somewhat handicapped in capture the dynamic nature of group interactions, since it does not intend to obtain data on actual interactions, but measures and manipulates the input and output variables, and merely produces inferences about what is going on in between. (McGrath & Altermatt, 2001:525.) Discussing group processes, social network analysis occurs as a proper choice to analyse group dynamics. It explains the degree to which network actors connect to one another and the structural setup of collaborative relationships. (Scott, 1992.) However, the data gained and analysed by general treatment of social networks has a somewhat similar disadvantage as the input-output, and questionnaire methods: it provides a snapshot, what is to say static information on group relations.
As a consequence, there is remarkably little scientific material on interactional group processes grasping the temporal aspect of team networks examined in terms of effectiveness and conducted by a well-established research that primarily focuses on actual development of real-time group interactions.
Therefore, the aim of the thesis is to present the changes of team networks primarily defined as sets of intragroup interactions from a temporal approach of the collaboration-performance nexus, thereby providing novel information on how team members use their relational resources across time of collaborative activity. The research question of the thesis is what kind of temporal network changes can be depicted by teams’ interactional patterns that promote quick solution finding. In other words, how team members use their social network, how it evolves over time of collaborative problem-solving.
Collaboration opens the way for attaining higher goals that are too demanding to be implemented by individual resources, thus, it can increase effectiveness and productivity. Therefore, collaborative teamwork tends to be pursued by successful organizations either to a strategic extent or at project level that implies a network of team members who are vigorously involved in a common activity. “Project teams have become a popular organizational form under circumstances that require coordinated actions directed towards a non-routine goal. They make up a relatively simple form of group to study, having a clearly-defined task focus and identifiable allocated resources.” (Rickards & Moger, 2000:273.) Therefore, project teams seem to be appropriate subjects of the thesis research to examine collaborative group activity. Despite its weakness, the input-output design of researches on group processes is frequently applied because the analysis of the process of actual ongoing interactions can be extremely time, effort and resource demanding. Nevertheless, the thesis intends to overcome this
4 difficulty to implement a complex research on real-time interactions of collaborative project teams, by revealing an easily accessible research field that provides an appropriate scene for social experiments, as well as a strong analogy with the organizational settings of project teams. Escape rooms are basically entertainment facilities offering playing fields where participants need to explore and communicate with each other. From a sociological point of view, escape rooms can be considered as laboratories, since the entire group process takes place in exactly the same, controlled environment for all teams under comparison. Moreover, real-time happenings can be constantly observed through a camera, while the game and successful task solution are based on collaborative group activity. Thus, the ‘laboratory features’ of escape rooms provide a field, where groups that are pursuing and performing mutual, cooperation-demanding goals can be followed, recorded and examined under controlled, equivalent circumstances, while the setup of the game accounting for time pressure under which teams have to explore and find a solution collectively supplies analogy with organizational conditions of project teams.
The game is executed on the following way: voluntary groups of people (2-6) get locked into a room, where different messages regarding the way out are hidden. The goal is well-defined and understandable to every participant: they have to exit the place within a one-hour time frame. For doing so, they must collaborate, since successful performance depends on cooperation. Under these circumstances, groups behave as teams, while members become interdependent by sharing a collective goal that requires high level of collaboration analogously to the labour characteristics of project teams in organizations. Moreover, the task itself is based on exploration and communication that are also central notions in case of project teams’ non-routine activity. Teams have to search, open locks and decipher codes. Neither of these tasks requires prior preparation, specific knowledge or abilities, therefore all players at least for the first occasion can be considered competent in problem-solving, just like the members of project teams who are selected according to competence in performing non-routine assignments.
Game activity can be constantly observed through three cameras (recording the scenes from different angles), which originally serve to prevent inadequate acts (e.g.: destruction of objects in the heat of the game), meanwhile their presence offers possibility to analyse real-time intragroup cooperation during project accomplishment.
Therefore, escape room provides laboratory circumstances to observe ongoing group interactions, furthermore, since the rules and characteristics of the game are analogous to those of project groups, an accessible and reliable research field is available in order to study how collaborative groups use their networks during project accomplishment.
5 One important aspect of team effectiveness, - also highlighted by the boundary condition of the game, - is the time during the given task becomes completed. The one hour time frame given to solve the problem accounts for the binary dependent variable of the analysis (success/failure), while occurring interactions during task fulfilment are primal units in the examination of temporal network changes which is the independent variable.
The central concept of the thesis concerns the interrelation between team performance determined by time spent on finding task solution, and temporal changes of team networks depicted and defined by interactions in order to investigate how collaborative team networks evolve over time, and how they correlate with effective project accomplishment.
The exploration of this relationship intends to provide a novel, temporal approach in managerial and organizational studies by utilizing a proper research field where conditions are given to objectively measure how interactional patterns of collaborative teams correlate with performance, that is what kind of temporal network changes can foster successful project accomplishment.
2. THEORETICAL CONTEXT
In this section I present the theoretical foundation of the research by reviewing the relevant scientific literature according to which two hypotheses are formulated.
The research intends to examine small groups’ interactional patterns by which their network structure evolves across project accomplishment. Social structure of communication is hereby defined as the indicator of the level and quality of collaboration which is determining in view of successful problem-solving. Besides, the given task to be achieved is also influential as its characteristics can define possible strategic considerations, thus interactional pattern that could favour to successful solution finding. Therefore, in the followings, I produce hypotheses on communicational patterns accounting for temporal changes of network structures from two directions. The first approach emphasizes that information flow is crucial in small project groups’ communication, while the second viewpoint originates in actors’ different network positions assumed to influence interactional patterns, and in the statement according to that distinct phases of exploration covers different network structures that manage to foster project-accomplishment.
6 2.1. Network structure and information flow
Members of project teams are collaborative problem solvers who explore and share ideas through intragroup communication. Intuitively, dispersion of communication between group members seems to be crucial in successful project-accomplishment.
Pentland (2012.) who investigated what makes a team successful claimed that the patterns of communication are “the most important predictor of a team’s success” (pp.4.). He found that successful teams share the characteristics that every member connects directly to one another, which implies cohesive, centralized group structure contributing to successful performance by promoting direct ways of communication and smooth information flow.
In social network science, the term cohesion refers to relations holding people together. (Bruggeman, 2008:12.) In our case cohesion accounts for communication providing intragroup information flow, thereby maintaining task focused collaboration. In small networks, density can be considered a proper measure of cohesion or connectedness providing a proportion of emerging and all potential ties that shows how strongly the network, in this case collaboration or ‘who-talks-to-whom’ graphs of small groups are connected.
Structural properties of naturally occurring networks are highly influential in behaviour and dynamics formation that is collective performance is significantly affected by network structure. (Kearns et al, 2006.)
Examining the relationship between structure and collaborative task performance, a recent research also reported that network groups significantly outperform equal-sized “collection of independent problems solvers” due to faster spread of information especially on good solutions in networks that are well-connected. (Mason & Watts, 2012.) These networks are also characterized by high level of collective exploration which is beneficial in respect of task-fulfilment. As a consequence, these networks are labelled as “efficient” ones, “where efficiency refers to the speed with which information about trial solutions can spread throughout the network.” (pp. 764.) In other words, network structure having high density promotes quick information flow, thus, surpasses “inefficient” networks with low density and independent exploration practice regarding collaborative task-solving.
Therefore, quick information flow claimed to be crucial in successful project-accomplishment by numerous social scientists seems to be promoted by network structure of small project teams having high density.
The reduction of average distance in the network that is high level of connectedness entailing small distance and high accessibility between individuals was found to have beneficial influence on collective behaviour. (Kearnset al, 2006:826.)
7 Talking about connectedness, the term usually covers two related aspects of networks.
Connectedness at structural level shows who is linked to whom, while connectedness in relation to behaviour refers to “the fact that each individual's actions have implicit
consequences for the outcomes of everyone in the system”. (Easley & Kleinberg, 2010:4.) In the thesis, networks are not handled as fixed social structures of relations between individuals, but as a set of emerging interactions in small project teams in a defined time interval of collaborative group activity. Therefore, connectedness in case of the research combines both structural and behavioural aspects, and primarily accounts for density, being a relevant measure of cohesion. Interactions as network ties necessarily display graph structures in given time periods, but also occur as basic units of collective behaviour, as they depict communicational dispersion determining the level of collaboration, hence, the outcome of teamwork based on cooperation.
Translating and matching these empirical findings into the temporal approach of the current research, according to the first hypothesis, I expect that highly connected that is dense networks characterized by homogeneous distribution of interactional activity across members and during the time interval of task-performance has a positive effect on collaborative solution finding. As a consequence, interactional patterns having balanced intragroup diffusion and being determined by low-variance in interactional distribution over time are assumed to foster successful problem solving.
2.2. Pulsing interactional pattern and exploration spaces
Collaborative task-fulfilment is the least likely to come to pass in a linear way, since different members of project teams having diverse ideas regarding distinct sections or aspects of the given project. In case of a team fails to process all the upcoming ideas, they may follow the strategy of task specialization in order to decrease time consumption. In this case, the network will manifest clusters that entail a disconnected overall graph structure in certain time periods of task-solving. At the same time, collaborative subgroups must integrate their partial results from time to time, otherwise successful project-accomplishment based on collaboration is doomed to failure.
Following the strategy of task-specialization, it manifests in temporal network of teams that can be visualized as a ‘pulsing’ interactional pattern, since it depicts variety in communicational distribution: duration of separated network components of specialized sub-groups with low overall density alternates with intervals of dense, completely connected temporal graphs while synthesis of partial results takes place. However, this implementation
8 of such an interactional pattern can be as destructive as it seems to be a great strategy without orderly executed integration of sub-findings, which is assumed to require either a coordinator, or a composed, well-organized and disciplined team that is able to collaborate smoothly. In order to investigate implications that may facilitate and supply conditions for the emergence of such pulsing interactional pattern, I refer back to the first hypothesis, where homogenous distribution of information flow implies dense temporal graphs, and it further entails network structures typified by similar contribution level of group members in collective activity, therefore temporal intragroup roles that are very much alike, thus members have similar amount of power and influence that account for analogously central positions in the temporal network. Actors are considered central in terms of having a greater influential power on the formation of intragroup interactions compared to other group members that justify prestige and personal importance. When salient roles, specifically leadership arises in collaborative teams, group dynamics, especially coordination issues come into view.
Leadership refers to “guidance of others in their pursuits, often by organizing, directing, coordinating, supporting, and motivating their efforts.” (Forsyth, 2009:246.) A leader has power to diffuse his/her ideas on the interactional pattern of collaborative groups, since such group members’ relational position is characterized by higher interactional embeddedness in the temporal network than any other actors’. Thus, the emergence of a leader necessarily modifies the interactional pattern by carrying out more interactions thereby causing an asymmetric dispersion of communication over time.
The presence of a leader may induce task specialization, while a member embracing role of the leader can be also enhanced by development of this kind of strategy. In this sense, leaders can have crucial, coordinating roles in collaborative teams where task specialization occurs. Generally speaking, „most people do not just accept the need for a leader but appreciate the contribution that the leader makes to the group and its outcomes.” (Forsyth, 2009:248.) A leader’s primary function is to control group focus in order to prevent tendency to stick at certain points of information share that are being time consuming, thus unfavourable in view of problem-solving. Numerous classifications and dimensions regarding leadership are detailed in scientific literature being significant in a sense of emerging disparities in productivity, efficiency and in group atmosphere. However, the thesis does not aim at investigating such features of group relations and focus on leadership, but it merely reveals its potential emergence and correlation with changing collaboration graphs in terms of density. Therefore, and due to the social network approach, leadership will be predominantly observed by the team members’ advanced nomination and by out-degree centrality measure.
9 Degree centrality can be considered as a measure of activity, since it shows all direct connections an actor has. Since temporal changes of human networks form the backbone of the research, accessibility of all members in case of small numbered teams during one hour intense collaboration is unambiguously expected. As for the quantity of direct reaching of fellow members is an opinion, degree centrality is an optimal measure for depicting influential and powerful actors of the given network.
Actors who possess high out-degree centrality are considered as opinion leaders, which term refers to “the degree to which an individual is able to influence other individuals’ attitudes or overt behaviour informally in a desired way with relative frequency” (Rogers, 2003, p. 27). According to the most general approach of leadership, leaders have greater participation than others, therefore, have more connections with other group members.
Concerning the interactional pattern, a leader’s systematic supervision can eventuate several time periods of problem-solving defined by disconnected, sparse overall networks interpreted as performances of subtasks in case of specialized subgroups.
Task specialization without periodic integration of sub-findings is supposed to obstruct the process of collaborative problem-solving due to erratic information flow, therefore being fatal in terms of successful task-fulfilment. A leader could bear important coordinating role in smooth alternation of the pulsing pattern’s two elements, however, leadership is not a necessary condition of systematic information share in a group.
Shore et al. formulated similar speculations on different network structures being beneficial during distinct processes of task-fulfilment by disassembling exploration process to the
investigation of information and searching for solutions, claiming that “network clustering has opposite effects for these two important and complementary forms of exploration.” (Shore et al. 2015:1432) Borrowing the terminology from the authors, exploration of information refers to the process of fact and data gathering being assumed to be important in task-solving; while exploration of solutions includes interpretation and dovetailing of facts in order to reveal answers and findings being relevant in successful performance. The former process is
demanded to be positively influenced by connected network structure, since information share tends to facilitate the search of particular information pieces explained by the importance of information flow. At the same time, increased connectedness elicits awareness of each other’s interpretation and theories, as a consequence, it is likely to induce convergence in team members’ mind-set that results in reduced diversity in knowledge, which thereby lowers the probability of effective exploration of solution space. “A high degree of group cohesiveness is
10 conducive to a high frequency of symptoms of groupthink, which, in turn, are conducive to a high frequency of defects in decision-making” (Janis, 1972, p. 199). (Forsyth, 2009:343). Shore et al. claim that cohesive network structure itself tends to trigger groupthink in solution space. In sum, they argue, that the same network structure can either promote or restrict task-solving, depending on whether the particular group process refers to information collection or to its interpretation.
Drawing a parallel with Shore et al.’s findings (2015.), the second hypothesis including the positive effect of a pulsing interactional pattern on successful performance of collaborative teams is considered further strengthened: as mentioned above, a pulsing network is characterized by alternation of duration of subgroups’ activity that represents low density in the overall network of the team, and periods of group-wide communication entailing completely connected collaboration graphs. The former process corresponds to the exploration of solution space when task specialization takes place in the subgroups of the certain team, while the latter part of task-solving coincides with the exploration of information space, when fact gathering, furthermore the integration of partial findings happens, too. Synthesis of sub-results can be also classified to the exploration of information category, since subgroups’ interpretation is transmitted as facts to other team members who are paying attention to different subtasks. Namely, the strategy of task specialization is only beneficial and reasonable in terms of performance if it minimizes time consumption of partial tasks. Therefore, it necessarily implies exclusive responsibility of a certain subgroup for a particular subtask; thus their findings cannot be inspected by other participants who are involved in other pieces of the project; otherwise the strategy of task specialization is not expedient. Regarding the pattern of temporal network use, if team members are consequent in sharing all the data collected in information space, and the exploration of information covers both the search for information and the synthesis of partial results (that can be interpretations disclosed as facts), I expect that the alteration of the two components of the pulsing pattern manifests in approximately equal duration of both types of network structures.
In summary, due to the non-linear feature of project to be accomplished, task specialization in collaborative teams’ activity can be a clever strategic consideration for time utilization. The interactional pattern that is induced by task specialization I named as ‘pulsing pattern’ since it entails the alternation of disconnected graph structures manifested by subgroups’ activity, and dense, group-wide networks of communication during integration of sub-results. The systematic changes of the two elements of the interactional pattern entail either a disciplined and well-organized group, or a team where social positions are likely to differ along actors’
11 influential power. The emergence of a leader bearing a salient central position in the group may favour to the implementation of such pattern by his/her coordination and supervision in group activity, but it can be also induced by the development of task specialization. However, leaders’ role is not an ineluctable condition to the formation of pulsing interactional patterns. Moreover, incorporating Shore et al.’s classification along exploration spaces, (2015.), the synthesis of partial findings is also considered as part of the exploration of information; therefore, pulsing interactional pattern is expected to manifest the change of exploration of information and the exploration of solutions, where the former one covers fact gathering and fact integration, while the latter one refers to theory making and interpretation. This pulsing pattern is assumed to consist of similar time intervals of the two network structure elements. Task specialization, - being in accordance with Shore et al.’s findings, - without a leader supposes a highly disciplined and well-organized but rather egalitarian team, where members manage to collaborate smoothly.
As a consequence, the second hypothesis states that ‘pulsing’ interactional pattern eventuated by task specialization corresponding to and manifesting different processes of exploration in fact and theory spaces tend to support successful performance by time utilization and well-organized collaboration that may also entail a leader’s potent coordination.
The formulated hypotheses identifying two disparate processes of network changes can be considered mutually excluding to a certain extent; but it does rules out the possibility of the detection of both interactional patterns as favourable types of temporal network in terms of successful project-accomplishment; however, due to the relatively small research sample I find it less likely to capture the appearance of both patterns.
In view of the conditions that could favour to successful teamwork, several compositional elements of social networks may prove to be relevant. However, since the primal target of the research concerns interactional patterns that account for temporal changes of team networks, selected descriptive characteristics of teams are handled as subsidiary variables being described in the next section, and later explained and related to the particular context through the analysis, to the extent that they provide meaningful correlation with project accomplishment.
12 3. RESEARCH DESIGN
In this section I present the foundation of the methodology framework: research field, method and approach of data collection, as well as the conceptualization and operationalization of key determinants in order to set the stage for seeking answers for the research question by the analysis.
3.1. Research field and method
The investigation of the independent variable that is the interactional pattern illustrating temporal network changes of naturally occurring, collaborative teams is revealed by covert observation conducted by video recording under unadulterated ‘laboratory’ and also ‘field experimental’ circumstances.
Escape rooms can be considered as laboratories for social experiments, since all group processes take place in the same, controlled environment in case of all participants. Moreover, time pressure under collaborative groups have to explore and communicate in order to find a solution makes the research field circumstances analogous to the organizational settings of project teams performing non-routine assignments. (In fact, escape room participants can be considered as actual project teams, as members are interdependent along a clearly-defined focus of collaboration-demanding task.)
Experiments are suitable to researches dealing with low number of and well-defined concepts and hypotheses. (Babbie, 2008:250.) It is especially appropriate to analyse small group interactions. Thus, experimental field of data collection is a natural choice of conducting the thesis research, since escape room environment provides precisely defined probative situations, where the relation of the independent and dependent variables are detectable, and reliable control of conditions is implementable ensuring reduction of unexpected impacts to a minimum.
Researches take place in escape rooms can be also considered field experiments to a certain extent, since these places are not artificially constructed mediums with the goal of conducting researches, but ‘naturally’ existing ‘playing fields’ being well-suited to data collection of group interactions. Moreover, covert observation implemented by video recording happens by the operators due to safety and legal considerations, which has to be officially accepted by all participants before the game. Therefore, this kind of data collection and the field experimental setting of escape rooms allow the elimination of the disadvantages of typical laboratory experiments, since the fact of the research observation is a subsidiary circumstance for
13 participants; thus, it is not likely to modify their behaviour because it does not elicit the perception of being watched and expected to behave in certain ways that favour to researchers’ anticipation. Thereby, the so-called Hawthorne effect that is behavioural modification of individuals occurring in response to their awareness of being studied by scientists is eliminated by covert observation and the field experimental settings.
The usage of cameras also facilitates a comprehensive and objective data collection where occurrence of heuristics being a typical disadvantage of retrospective reports is excluded, and collection of fine-grained data is implementable easily. Moreover, video recording also eliminates inaccurate, implicit assumptions about the actual group interactions that accounts for a primal weakness of the input-output research design. Electronic recording systems permit a delayed and thoroughly implemented data coding, since videotapes allow to repeat given action series, therefore observers are not constrained to make real-time, thus potentially less grounded or premature judgements and data selection. (McGrath & Altermatt, 2001:530.) As a consequence, the chosen research field, data collecting technique and the method situated between laboratory and natural experiments provide a proper layout to observe collaborative team activity during project accomplishment; in addition, it supplies necessary time and opportunity to gather and code sheer and objective, fine-grained data on interactional patterns in order to investigate how temporal network of project groups evolve over time and how it correlates with successful problem-solving.
3.2. Approach and data preparation for the analysis
In the first place the independent variable, namely the team's network is scientifically determined by the presentation of its analogous and variant, hereby called “reinterpreted” handling of social network approach.
In social networks, ties can represent friendship, business and any different kind of relationships. Due to the independent variable of the thesis being team networks, social network perspective occupied with relational characteristics clearly denotes the starting point of the disquisition. At the same time, social network data provides static information on relationships along different dimensions represented by ties that occur between actors. For instance, in case of workgroups, network data can supply information on team structure, hierarchy and social roles that relations and features are relatively longer-standing ones. In organizational and managerial studies these information account for one of the primary contributors in the explanation of team performance, since structural and compositional information of teams necessarily influence the outcome of collaboration; however, they do not
14 supply direct, explanatory power for actual processes taking place in collaborative groups across task-solving. Consequently, general application of network analysis is adequate, but bears limited applicability in view of temporal network changes due to the static nature of evidence it meant to provide. Thus, besides the obvious utilization of network perspective as the backbone of the research is self-evident, grasping temporal aspects of ongoing team activity requires a modified interpretation and method of mainstream social network analysis on the one hand, and on the other hand it entails a more complex analysis that combines but extends beyond general treatment of social networks.
In the research, networks are not handled as fixed, static relations of teams, but as sets of actually occurring interactions within a well-defined time period during which teams perform a collective task. Therefore, interactions correspond to ties, moreover, in this case they refer to social aspects of individual contributions, participation in collaborative activity, thereby we can refer to them as collaboration or ‘who-talks-to-whom’ network ties made of intragroup communication. In conclusion, occurring communication between team members during collaborative task performance is the basic unit of network changes over time.
3.2.1. Interactions as network ties
Interactions taking place during collaborative team activity are considered as ties between participants, thereby determines the collaboration graphs, network structures of project teams. Interactions are hereby defined as all explicit verbal behaviours, communication of group members that they perform in relation to one another (and the responses they are giving in return), focusing on the task they are accomplishing over time.
Information on team activity is recorded by cameras that provide a comprehensive dataset, however data obtained and being coded is inevitably selective; cannot and do not even intend to incorporate all potential aspects of group behaviour. (McGrath & Altermatt, 2001:532.) As a consequence, central data of the research that is network ties are pair-wisely coded interactions with the denotation of involved actors anonymously: the sender that is the source and the receiver is regarded as the target of the interaction.
The unit of activity is defined in temporal intervals of one minute. Minute-based, equally long phases of aggregated communication are the smallest interpretable units in terms of temporal networks manifested by emerging real-time interactions. Moreover, natural time periods such as chronologically coded sequences are in accordance with the technically feasible implementation of the time-and effort-consuming research topic. Within these units all the
15 occurring interactions are coded according to the “source-target” scheme. (McGrath & Altermatt, 2001:535.)
The activity’s volume is determined by its temporal extension: the duration of the appearing interactions are classified into long and short ones, where the former category covers about twenty seconds or longer periods of uninterrupted communication, while the latter one logically denotes shorter intervals.
"time" "send" "rec" "dur"
21:32 E ME S 21:32 E D S 21:32 E MA S 21:32 D MA S 21:32 D ME S 21:32 MA ME S
Hereby, basic parameters of coding of groups’ interactional information gained by the primal data collecting tool of video records are established and prepared for the provision of a team’s collaboration network and for its statistical analysis.
3.2.2. Dependent variable
Performance, as the dependent variable of the research is primarily defined by the speed of solution that can be categorized into a binary variable, since a determined boundary condition of one-hour time frame is given by the research field environment. Thus, teams that do not manage to exit the place within an hour by effective collaborative activity are automatically considered unsuccessful groups, and labelled as “0”, while teams conducted a successful project accomplishment are denoted by code “1”.
Participants are holding a walkie-talkie during the game by which they are able to ask for the operator’s guidance if the team gets excessively stuck. Help request, however is a double-edged issue to be interpreted: it is tempting to consider them as explicit signs of periodic failures in task-fulfilment, thus, treat them as errors that could form an additional measure of task performance. Although, at the same time, help-seeking may account for a clever way of
16 available information exploitation, as well. Therefore, the number of help request is cannot be considered as reliable indicator at this point, thus it is not involved in the analysis as a supplementary dependent variable. Nevertheless, the number of help requests is examined as a separate indicator in relation with performance in order to gain answer for the aforementioned ambiguity pragmatically.
3.2.3. Research sample
Based on operators’ opinion from three different escape rooms during field trials, the optimal group size for collaborative task solving under such circumstances is 4-5 people.
In case of escape rooms, the possible wider range of participants (2-6) seems to be a business consideration. Therefore, and in accordance with reasonable deliberation regarding feasible implementation of the thesis, the research covers 16 naturally occurring groups with the particular size of 4-5 people.
In order to perform the task, and exit the room, teams have to explore and communicate during searching, opening locks and deciphering codes. Neither of the tasks requires prior preparation, specific knowledge, skills or abilities officially, however, recurrent participants can be logically expected to perform the task quicker than first-timers. Therefore, the research sample consists of 16 teams where all members participate in the game for the first occasion, thereby all of them are considered equally competent, which selection is implemented in order to reduce unexpected external impacts that could undermine reliability of the research.
The game is suited to participants from the age of 12, however, younger players might be lack of associative capabilities and explorative practices compared to older ones. Moreover, such young participants are likely to arrive and form a group with their parents, which entails specific intragroup relations that can unforeseeably influence the interactional pattern. As a consequence, I controlled the sample for groups consisting of (relatively) adult members, thus the observed participants’ age varies between 17 and 45 years.
3.2.4. Additional data collecting tool and further analysis plans
In spite of the fact that the task does not require specific knowledge and abilities, information based on preliminary field trials and experiences of escape room operators suggest that certain kind of social skills can function as advantages. These specified skills and abilities that operators found prosperous (being consistent with the hypothesis) are: capacity to pay attention to each other that manifests in distribution of interactions between team members
17 providing information flow and high level of collaboration through constant communication; and team spirit that facilitates active participation and constant focus on the task. The development of team spirit and paying attention to each other require balanced cooperation between group members that could be redounded by prior team life.
Therefore, - although the network data gathered by video recording provide relational information that can be analysed by cohesion and centrality measures, - a questionnaire as an additional data collection tool is designed to make inquiries on a few compositional elements being potentially influential to the interactional pattern of teamwork.
The first item of the survey attempts to investigate basic group relations beforehand. The question asks the participants to “have their current team mates sat down” in a minibus for a one-day bus trip; once with the denotation of the bus driver selected from themselves, and once without a driver. The chosen person of the driver is assumed to be a central or leading character in the team. At the same time, it is interesting whether the nominated individual (pre-established leader) corresponds to the person who is objectively measured to have a central position in the group regarding his/her participation level and power to influence (emerging leader) displayed by out-degree measure.
The second item of the survey investigates how long and from where the team members have known each other, which information may have an impact on the formation of team spirit, moreover the duration and source of prior acquaintance might also influence distribution of communication between the members involved. Besides this item, the relational data also allows to examine entropy in this question, that is how (un)evenly team members diffuse their communication across their fellow participants during task performance.
The survey is constructed to investigate permanent and relatively palpable group connections beforehand, as well as recording basic demographic data on age and gender. Although as it is stated above, successful accomplishment of the game does not require any kind of specific skills or prior preparation, certain kind of occupational background may emerge as advantage. For example, individuals with military experiences, or with jobs and working environments where coping with and staying focused under highly stressful situations are basic requirements, may tend to have mental advantage during performance. Therefore, another element of the questionnaire asks the participants to report on their current job. (However, it is questionable how this personal, potential advantage can be translated for the benefit of the team).
The additional questionnaire as a secondary data gathering tool is designed to supply information on compositional and structural variables of teams that might contribute to
18 successful performance, furthermore, it also permits to compare and contrast previously declared, perceived nature of intragroup relations of team members with the relational network data gained by the primal data collecting method of video records on group interactions.
Questionnaire data of age, gender, leadership and duration of prior acquaintanceship are treated by correlation and regression models in relation to the binary dependent variable of the analysis.
Besides the characteristics and coding of network ties, a bunch of relational network data is also available for analysis. Team spirit that might be manifested by intragroup tie distribution is also examined by entropy measure, besides the possible influencing compositional data of duration of prior acquaintanceship. Reciprocity measure providing the proportion of requited and all ties as a potential indicator of the extent to which team members pay attention to each other is also analysed, as well as centrality measure of out-degree to reveal the “emerging leader” besides the “pre-established” one recorded by the questionnaire.
Short and long ties and their proportion are not tightly-connected to the examination of temporal network, thus they are investigated separately, also in terms of performance to explore if duration of communication has an impact on successful project-accomplishment. Moreover, potential correlation of number of help requests with performance is analysed to answer whether they can be considered errors, or a clever way of information exploitation.
I divided the data analysis into three parts. The first two sections form the backbone of the research and the hypotheses testing. Firstly and most importantly, I present the central part of the analysis, namely, the temporal network changes in relation to the binary coded dependent variable (success/failure). In this section I only focus and use data that account for temporal alternation of graph structures that I capture by a density measure. Since basic units of the independent variable are emerging interactions, an important implication of communication is its frequency. At the same time, since interactional intensity does not have primal explanatory power in changing network structures, I investigate potential clustering of communicational frequency along successfulness in the next section of the analysis by dynamic time warping method. The last part of the analysis includes the treatment process of other variables such as the compositional data collected by the questionnaire on the one hand, and the relational information on the entire network of project teams gathered by the primal data collecting video records measuring cohesion and centrality implications, on the other hand.
19 4. ANALYSIS AND RESULTS
I analysed the research sample consisting of 16 teams - in case of which the ratio between successful and those who could not perform the task in time is fifty-fifty percent, - by R statistical programming environment. Applied R packages include igraph; dtw; qqplot2; dplyr and reshape2.
I investigated network changes by density measures of temporal graphs quantitatively, and also handled them as stylized facts to visualize and present changing graph structures in the most outspoken way. The examination of interactional intensity as a tightly bounded aspect to the research focus but still considered as a secondary factor was analysed by dynamic time warping technique in order to reveal potential clustering of interactional frequency along successfulness. Furthermore a number of correlation, regression models, and non-parametric Wilcoxon rank-sum tests were used during broadening the scope of the research by the analysis of compositional variables gained by the summarization of the survey data and relational variables of network properties.
4.1.Temporal network changes
In this section, I present the analysis of temporal network changes in the frame of the investigating process of the research question that how collaborative teams use their networks over time and how it correlates with successful performance.
Two interactional patterns were formulated and expected to positively influence task-solving: one predicts frequently high and relatively homogeneous temporal distribution of intragroup interactions; while the other one depicts a pulsing pattern where intervals of sparse graphs taking place due to task specialization alter with dense graph structures covering periods of the synthesis of sub-findings.
In order to conduct an analysis on temporal network changes in terms of time spent on performance, first I needed to select the most outspoken indicator by which network structure can be captured and quantified in a way that is informative at the presentation of temporal changes.
Talking about interactional pattern, it implies two primal characteristics of communication: its dispersion between team members, and its frequency. At the same time, interactional pattern is observed in different project teams from a social network approach with the aim of exploring how networks change during time of task performance. Therefore, interactional patterns as sets of emerging communication can be considered as the ‘raw material’ of the
20 independent variable that is temporal network changes. Thus, in the first place I had to focus on structure by which networks are typified the best in this case. Hence, referring back to the two main traits of interactional patterns in terms of small networks, density seems to be a more adequate measure bearing explanatory power on networks structure, in contrast with interactional intensity as a quantified number of occurring ties provides no information on interactional diffusion, communication between team members by itself. However, I also experimented with combining the two aspects of communication into one indicator that considers both dispersion and the number of interactions as multiple ties emerging in the network. This experiment resulted in values much higher than the maximal density value of 1. Thus, getting numbers as 4.665 for example, proved to be not informative either on graph density or communicational frequency, since we cannot interpret none of these measures clearly. These values may cover completely connected graph structures with lower interactional intensity, but they can also inform about a sparse graph with high frequency of communication within connected network components, different subgroups. Hence I opted for the separated analysis of network density and interactional intensity to maintain clear and reasonably interpretable measures. Besides, as we focus on networks, I considered density as the most adequate indicator displaying changing structures of project teams.
Thus, I grasped the structure by establishing density indicators that manifest graph connectedness.
Since the data on time of group activity is coded minutely, as an initial investigation I worked with minute-based contact ratios (MBACR) defined as the proportion of participants that team members have an interaction with on average in each minute of task performance. This approach provided 16 data series of teams determined in consecutive minutes, that I illustrated by probability density function, to reveal and depict well-visible discrepancy between the teams’ network structures according to successfulness.
Subsequently, I applied moving window technique to explore network structure discrepancies in different time frames by leaving behind the minute-based determination, and observe average contact ratios in varying time units. I visualized temporal changes according to the moving window defined to five minute on graph density plots to illustrate and highlight disparities in network changes of successful and unsuccessful teams.
The dependent variable that is time spent on performance is considered as a binary category variable: successful teams that solved the problem in one hour are coded as “1”, while teams that went beyond the time frame are denoted as “0”.
Both of the used minute-based contact ratio (MBACR) and average contact ratio (ACR) calculations correspond to the density formula of undirected and directed networks; however,
21 I found it proper to present a more detailed process of the establishment of these variables in order to better illustrate my choice of applying density measures regarding the analysis of network changes. Hereby I demonstrate the process of examination step by step by using the data of Team 1, as an example.
As a recap on the coding schema, the table shows that besides time is determined in minutes, the data gained from the video records on group activity was coded with the indication of the source and target, (both coded anonymously), and the duration of the communication, where “L” denotes long interactions defined in twenty seconds or longer intervals of uninterrupted communication; and “S” accounts for all the short communication took place below twenty seconds.
"time" "send" "rec" "dur"
21:32 E ME S 21:32 E D S 21:32 E MA S 21:32 D MA S 21:32 D ME S 21:32 MA ME S
The graph of Team 1 in the first minute of group activity gained from the coded information looks like this:
22 First of all, I examined in every minute that how many other actors, the given team member has an interaction with. In each minute, an interaction counts only once; its directivity and duration as potential weights along which tie strength could be demonstrated are not taken into account at this point. As the first phase of the analysis my aim was to capture and present network structure in the most basic way in accordance with the coded data to have an initial picture on graph density on which I can build further and sophisticate the analysis by increasing its complexity.
Thus, in case of Team 1, D actor had interactions with 3 other participants; therefore, 3 interactions were counted and considered as the contact number of team member D, as the second row of the first, binary symmetric contact matrix shows.
E D MA ME E 0 1 1 1 3 D 1 0 1 1 3 MA 1 1 0 1 3 ME 1 1 1 0 3 3 3 3 3 E D MA ME E 0 0.333 0.333 0.333 0.999 D 0.333 0 0.333 0.333 0.999 MA 0.333 0.333 0 0.333 0.999 ME 0.333 0.333 0.333 0 0.999 3.996 0.999 0.999 0.999 0.999 0.999 3.996 0.999 0.999
By multiplying the symmetric contact matrix by the weight vector that I implemented to normalize the matrix by taken into account the number of participants, contact numbers became ratios showing which proportion of the team, the given actor has an interaction with, in the given minute denoted by the second matrix. Thereafter, I averaged these results to gain the ratio of the team that members were in interaction with on average. The value of 0.999 in case of Team 1 indicates that they had a completely connected, maximally dense network structure in the first minute.
23 I computed these contact numbers of group members for the entire team and for every minute of problem-solving activity. By this treatment, I received a ratio value in each minute of all teams that shows which proportion of the team, group members have interaction with on average. This value as a quantified measure of how dense the minute-based graph structures corresponds to the density formula of undirected networks that is m/[n(n-1)/2], where m denotes the number of ties, while n indicates the actors.
In case of Team 1, the gained data series can be depicted on a line plot, where the horizontal axis shows time of group activity defined in minutes, while the vertical axis shows the values of minute-based average contact ratios (MBACR) accounting for graph density.
Then, I examined how these MBACRs correlate with successfulness. For doing so, I used probability density function to achieve information on the probability of average contact values’ incidence according to successfulness. This function provides information on possible differing range of MBACRs at successful and unsuccessful teams. Thus, the shape of the distribution curve bears information on that to what extent graph density manifested by MBACR influences successfulness.
24 Unsuccessful teams’ MBACR values are presented by the red curve with code 0, while teams that solved the task within one hour are labelled as 1, and visualized by the blue curve. Thus, the probability density function demonstrates the relative likelihood of different MBACR values’ occurrence according to successfulness, so the plot gives an initial picture about how these values in the two team categories relate to performance. The horizontal axis indicates MBACR values, while the vertical axis shows the probability of the occurrence of these values.
The plot shows relatively disparate distribution of MBACR values’ probable occurrence in the two (successful/unsuccessful) team categories.
At successful teams (1), the average contact value is somewhat higher than it is at
unsuccessful teams. Therefore, although the difference is not salient, the initial picture on
teams’ network changes inform us about that in successful groups, participants, -
proportionally to the number of team members, - tend to have interactions with more of their fellow members on average than it is the case in project groups who failed to complete the
Figure 4: Probability density function on MBACRs of all teams according to successfulness
25
task. As a consequence, the results so far suggest, that in the examined sample of project teams, successfully performing ones incline to have denser network structures during collaborative group activity, than unsuccessful ones.
Observing temporal network changes in relation to the minutely defined dependent variable was the first step in the analysis, since ‘leaving’ one minute for the development of communication can be seen as a ‘strict’ limitation in exploring varying interactional patterns along successfulness. At the same time, unit of activity was determined in minutes, equally long time intervals of communication, because this is the smallest interpretable unit of temporal networks manifested by occurring real-time interactions. Moreover, this coding schema is in accordance with the feasible implementation of the research topic. Nevertheless, I did not want it to restrict the volume and complexity of temporal analysis. Therefore, as the next phase of the examination, I applied moving window method to explore differences of graph density in different time frames along successfulness. By this method, the values of the 16 teams’ data series can be analysed in units (windows) determined by different time lengths, herewith defined in window widths of 1-2-3-4-5 minutes. Using this approach, I employed different, bigger temporal resolutions than one minute for the examination. Thus, this technique can be explained as the extension of barriers of the minute-based time units, and ’providing more opportunities for teams to evolve their communication’ by observing their interactions in wider time frames.
Again, I captured network structure by primarily examining that how many other actors, the given team member had interactions with. For doing so, I made the contact matrix for each of the applied time window where interactions between actors were described with the denotation of tie direction.
Thus, in case of a 4-membered team, I made an asymmetric 4x4 matrix, where ties are indicated by a binary code: their absence is coded by “0”, presence is marked as “1”. Moreover, to value communication, I made a distinction between reciprocated and unrequited ties by considering the latter one as a half-valued interaction, while requited communication was counted as a full value tie. A directed interaction has a source and a target actor. When there is another directed tie following the previous one between the same two actors but with reversed role of the source and target of the interaction, we can talk about a reciprocated tie; thus, a requited interaction is made of two directed ties with opposite directivity between two actors. Hence, if we consider a reciprocated interaction one unit, a directed tie logically bears proportionally half-value compared to the requited one.
26 I opted for this treatment of using binary codes and incorporate reciprocity, because I needed to capture a primal indicator by which network changes can be observed that is the graph’s communicational, density structure: the occurrence and distribution of ties between actors. Duration of interactions (short/long) is examined separately, since it is not relevant in view of connectedness; instead, I involved reciprocity to indicate a kind of tie strength or interactional quality. The number of emerging interactions per minutes accounts for frequency, that I found rather adequate to implement in the context of interactional intensity than involve it in a cohesion measure.
Thus, in order to grasp network structure through density, the number of team members also had to be incorporated to the contact matrix denoting the presence or absence of weighted interactions according to reciprocity. Therefore, as before, the same weight vector was introduced, which is practically 1/(number of team participants of the given group-1).
The figure below shows the graph structure of Team 1 in the 5 minutes customized time frame from the second minute of problem-solving.
Again, I coded the relations represented by interactions between team members to a binary matrix where I have also taken into account tie directivity.
Figure 5: Graph structure of Team 1 in the 5 minute time frame starting from the second minute of task performance
27 E D MA ME E 0 1 1 1 3 D 1 0 1 1 3 MA 1 0 0 1 2 ME 1 0 1 0 2 3 1 3 3 E D MA ME E 0 0.333 0.333 0.333 0.999 D 0.333 0 0.333 0.333 0.999 MA 0.333 0 0 0.333 0.666 ME 0.333 0 0.333 0 0.666 3.33 0.833 0.999 0.333 0.999 0.999 3.33 0.833 0.833
Normalizing the asymmetric contact matrix by the weight vector, the sum of the rows shows which proportion of the team, the given actor sent an interaction to, in the given time frame, while the sum of the columns indicates that which ratio of the team, the participants received a tie from. Thereafter, I averaged these results by the number of team members in order to gain the ratio of the team that members were in interaction with on average either as the sender (rows) or the receiver (columns) of communication. Hence, I gained the average contact ratio (ACR) that characterizes the graph density of the entire team in the given window, which is 0.833 in case of Team 1, as the implication of tie directivity resulted in the same value of 0.833 considering the rows and columns of the contact matrix. This measure corresponds to the density formula of undirected ties: m/n(n-1).
Thus, the established variable I referred to as average contact ratio (ACR) informs about teams’ graph structures through density the best, by showing which proportion of the group, team members had an interaction with averagely. The ACRs were calculated in case of all teams in the presented way in 1-2-3-4-5 minutely defined time frames (windows).
Subsequently, I examined to what extent these results could explain successfulness.
An important aspect here is that the 16 data series of teams have different lengths since they accomplished the task within distinct time intervals, thus applied time windows can be placed on them variably. I found it a potential misleading factor. Moving window method calculates
28 average values of the available data series of groups. Since these data series are longer in case of unsuccessful teams due to longer time of task performance, the computed averages in different time frames are based on more observations, which fact unambiguously distorts comparability, thereby causes unsubstantiated results. Thus, I inspected how long data series remain by using different window widths, and I based the analysis on the least common multiple of all teams which is forty minutes problem-solving activity conducted by the quickest team.
In general, I can state that the wider the time windows were slid per minutes on the forty minutes data series, the more visible the differences are between the distribution of successful and unsuccessful teams’ density values.
The difference between average network density of successful and unsuccessful teams using the 1 and 2 minute “sliding windows” were unambiguously low to make far-reaching deductions.
In the three minutes window, however, probable occurrence of ACR value showed visible discrepancy between successful and unsuccessful teams.
Figure 6: Probability density function on teams’ graph density according to the three minutely defined moving window
29 Even sharper difference can be seen between the team categories according to the analysis conducted by the four minutes made window unit.
In the moving window customized for five minutes, data curves indicating the probability of ACR value emergence in the two categories inform about explicitly much higher density values of successful teams’ networks than ACR values occurring in unsuccessful teams’ five minutely defined graphs.
Figure 7: Probability density function on teams’ graph density according to the four minutely defined moving window
