Once Upon a Time: The Process of Structure in MTSs Roger Leenders, - - PowerPoint PPT Presentation
Once Upon a Time: The Process of Structure in MTSs Roger Leenders, - - PowerPoint PPT Presentation
Once Upon a Time: The Process of Structure in MTSs Roger Leenders, Noshir Contractor, Leslie DeChurch A Network Approach to the study of MTS Antecedents? Consequences? A Network Approach to the study of MTS Antecedents
A Network Approach to the study of MTS
Antecedents? ¡ Consequences? ¡
A Network Approach to the study of MTS
Antecedents ¡
- Informa3on ¡need ¡
- Loca3on ¡
- Team ¡culture ¡
- Task ¡dependence ¡
A Network Approach to the study of MTS
- * ¡Some ¡teams ¡
perfectly ¡located ¡to ¡ be ¡innova3ve, ¡but ¡ not ¡to ¡be ¡efficient ¡
- * ¡Some ¡teams ¡
perfectly ¡located ¡to ¡ be ¡efficient, ¡but ¡not ¡ to ¡be ¡innova3ve ¡
- * ¡Others ¡good/bad ¡at ¡
neither ¡MTS ¡ performance ¡low ¡ nodes: teams, tie: knowledge sharing
“Classic” Social Network Analysis
Data requirements: single (aggregated) observation of a network
Adding change: MTS Network Dynamics
Behavior
Longitudinal Social Network Analysis
Most common: Markov chain + agent-based model For example: “network change can be explained by a tendency toward reciprocity, transitivity, and a preference for similarity in work commitment.”
Longitudinal Social Network Analysis
Data requirements: several observations of a network in a meaningful state Useful for: modeling how networks develop through a series of states
Taking MTS Temporal Dynamics a Step Further: From Snapshots to the Movie
* Increasingly we have access to (full) event data (e.g., sensor data, server logs, videotaped interaction) * Temporally binning data is a waste of beautiful data * The world changes in real time and step-by-step, so much is gained by modeling it that way * There is much variance to be explained in MTS performance, modeling the movie allows us to open the box for many new explanations and more detailed analysis
event e = (sender, receiver, weight, time) The event sequence (e1, . . . , ei−1) determines event network Gi
Δte = time difference between two given events f(e│Ge; θ) = fλ(sendere, receivere, Δte│Ge; θ(λ)) * fµ(we│sendere, receivere; Ge, θ(µ))
A little bit of Greek
probability density that the i’th event happens after a waiting time of Δte for given sender and receiver—given the state of the network Ge and the rate parameters θ(λ). probability density that event e has weight we—given the network Ge and values for the weight-parameters θ(µ) and given that the next event involves ae as source and be as target.
The essence of the relational event model
The probability of event ei (given Gi) is modeled so that the estimated parameters show: * which properties of the sequential structure of past events increase/decrease event rate (frequency)? * which properties of the sequential structure of past events increase/ decrease event weight (magnitude/type)? The model allows for parallel development of event networks and individual states (e.g., co-development of network ties and satisfaction) and can be applied to small and large multiteam systems.
→ originators: Butts (2008), Brandes et al., (2009)
How to model event rates between A and B: some variables
Habitual inertia
HISTORY FUTURE
Hypothesis: The rate of communication of A to B increases with the past volume of communication from A to B
TIME
A B A B
How to model event rates between A and B: some variables
Volume-Induced Reciprocity
HISTORY FUTURE
Hypothesis: The rate of communication of B to A increases with the past volume of communication from A to B Some variations: reciprocity within a team, reciprocity between roles, reciprocity between teams
TIME
A B A B
How to model event rates between A and B: some variables
Interteam Mimicry
HISTORY FUTURE
A C Hypothesis: The rate of communication of A to B (on another team) increases with the past volume of communication from A's team mate C to B B
How to model event rates between A and B: some variables
Broker skipping
HISTORY FUTURE
A C Hypothesis: The rate of direct communication of A to B increases with the past volume of indirect communication from A to B through another MTS member B
How to model event rates between A and B: some variables
Shared Value Induced Communication
TRUST FUTURE COMM.
A C Hypothesis: The rate of communication of A to B increases with the similarity of their trust vis-a-vis other MTS members B
How to model event rates between A and B: some variables
Dependence-based communication
TASK DEPENDENCE FUTURE
A B Hypothesis: The rate of communication of A to B increases with A's task dependence on B
- MTSs ¡performed ¡a ¡humanitarian ¡aid ¡task ¡based ¡on ¡the ¡real-‑3me ¡strategy ¡
game, ¡“World ¡in ¡Conflict”. ¡ ¡
- Two ¡cross-‑func3onal ¡teams ¡worked ¡to ¡ensure ¡bordering ¡regions ¡were ¡
passable ¡by ¡a ¡humanitarian ¡aid ¡convoy ¡protected ¡by ¡US ¡and ¡UN ¡teams. ¡ ¡
- Each ¡of ¡the ¡two ¡teams ¡worked ¡to ¡ensure ¡their ¡respec3ve ¡regions ¡was ¡safe ¡
from ¡enemies ¡in ¡the ¡areas ¡where ¡the ¡convoy ¡would ¡travel. ¡The ¡teams ¡ needed ¡to ¡coordinate ¡with ¡one ¡another ¡in ¡order ¡to ¡gather ¡and ¡share ¡ intelligence, ¡follow ¡rules ¡of ¡engagement, ¡and ¡neutralize ¡enemy ¡hotspots. ¡
- Each ¡par3cipant ¡was ¡seated ¡at ¡an ¡individual ¡PC-‑worksta3on ¡and ¡wore ¡a ¡
microphone-‑equipped ¡headset. ¡
- The ¡US ¡team ¡and ¡UN ¡team ¡were ¡located ¡in ¡different ¡rooms. ¡
A small illustration (preliminary)
A participant communicates with his teammates to coordinate clearing a hazard to the convoy. All communication between team members is recorded.
MTS Platform – Simulation Interface
Zone ¡shading: ¡ Red ¡– ¡Insurgent ¡ Blue ¡– ¡IED ¡ Green ¡– ¡Safe ¡ Object ¡Loca3ons: ¡ Purple ¡square-‑ ¡ convoy ¡loca3on ¡ Cyan ¡square ¡– ¡ player ¡loca3on ¡ Red ¡square ¡– ¡threat ¡ loca3on ¡ Yellow ¡square ¡– ¡ally ¡ loca3on ¡
Flow of the Experiment
Training ¡ Prac3ce ¡ Actual ¡Mission ¡ Planning ¡Episode ¡ Performance ¡ Episode ¡ Phase ¡I ¡ (15 ¡mins) ¡ Phase ¡II ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Phase ¡III ¡ (20 ¡mins) ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡(20mins) ¡
Some preliminary findings
Overall: * habitual inertia is high: communication is self-reinforcing, follows clear self- reinforcing pattern * reciprocity is high: the more A has sent messages to B, the quicker B will respond to A * Interteam reciprocity is negative: this balances against general reciprocity: quick response to team members, but response speed is much lower when the messages come from a member of another team
MTS reconfiguring itself
Some teams were allowed to communicate freely among each other, but reconfigured themselves into a chain-like structure. What happened: * interteam mimicry became negative: increased efficiency in interteam communication patterns—increased time until A sends a message to B when C already does this a lot
A C B A C B
MTS reconfiguring itself
* interteam inertia remains: the frequency and timeliness of communication between the teams is still maintained, but organized through informal role- specialization (boundary spanners)
A B C D Phantom Stinger Completely Connected Communication A B C D Phantom Stinger Chain Communication
Performance: aggressive
Teams that score high on “aggressive” performance show a differential tendency for: * broker skipping (negative) * common allocator (positive)
A C B A C B if the info is getting to B already, A won't spend his time on doing that as well If C gets his info from both A and B, then better send it to B directly
Future directions
* generalize to multiplex networks:
- - media switching
- - one kind of relationship triggering another
* further integrating event network dynamics with emergent states * develop additional event structures * study larger MTS's * use findings from this type of analysis to further time-based theory
Klik ¡om ¡de ¡opmaak ¡van ¡de ¡
- verzichtstekst ¡te ¡bewerken ¡
- Tweede ¡overzichtsniveau ¡
Derde ¡overzichtsniveau ¡
- Vierde ¡
- verzichtsniveau ¡
Vijfde ¡
- verzichtsnivea
u ¡
Zesde ¡
- verzichtsnivea