Communication, Shared Mental
Models and Team Performance
Robert Kraut
Susan Fussell
Javier Lerch
Carnegie Mellon University
Coordination is the Glue for
Teamwork
• Coordination = Extra effort multiple agents
must provide to achieve a goal, above what
they would need if they were working
independently
• Techniques for coordinating
– Division of labor
– Communication
– Shared models
What Roles Do Mental Models
Play in Coordinating?
• Mental model
– Mental representation of some dynamic process or
system
– One type is knowledge of teammate’s competencies =
transactive memory
• Mental models can improve coordination by:
– Substituting for direct communication
– Allowing effective task assignment
– Providing common ground for communication
efficiency
– Improving project planning and execution
• May lead to improved performance
Secondary question:
What leads to shared mental
models?
• History together => opportunities to
observe
• Communication
– More communication
– Evenness of communication
• Division of labor
– Read expertise from roles
Method
• Setting: Management game
– 50 teams manage a simulated consumer products
company over 14 weeks & 2 simulated years
• Data collection
– 3 waves of questionnaire administration
– Evaluations by external “board of directors”, based
on plans and performance
– Firm price based on stock market
• Analysis
– Predicting changes in coordination & outcomes
– Panel design, using mixed model
• Team as a random factor
• Auto-regressive error structure
Model and Hypotheses
•Predictions
:
Experience together, more communication, and more even communication
will enhance development shared models
Communication and consensus will influence performance through
coordination
Communication and consensus will have their beneficial effects early
Communication and consensus will substitute for one another
Communication
& structure
Shared
models
Task
process
Outcomes
History
Communication
Amount (mean)
Evenness
(reversed gini)
Functional expertise
Amount (mean)
Distribution
(reversed gini)
Agreement
about who
knows what
(mean r)
Coordination
Board
evaluation
Stock
price
Important Measures
• Communication
– Volume = Mean amount of pairwise communication
– Evenness = Reverse gini coefficient on volume of
communication
• Shared mental models = Consensus on who knows
what
– Mean correlation of members’ assessment of each others
knowledge of marketing, finance and production
• Coordination = Multi-item, self report scale
– E.g., Each member of my team had a clear idea of the team's
goals.
– E.g., Tasks were clearly assigned. I knew what I was supposed
to do
• Performance
– Stock price
– Evaluations by boards of directors after review
Summary of Results
Amount
Communication
Shared
Model
Coordination
Stock
Price
Board
Evaluations
Division of
Labor
History
Group
Expertise
Evenness of
Communication
.48
-.33
.62
.05
.11
.41
.38
-.32
.14
-.38
.31
.78
Standardized beta weights
All dependent variables control for lags => measurement of change
Only relationships p < .06 have been retain
-.14
Predicting shared mental models
• Shared mental model= Shared person
perception
• Average correlation among team members about
how much each person knows about finance,
marketing, production
• Predictors:
• Passage of time - No
• Communication
– Amount - No
– Evenness - Yes
• Division of expertise- Yes
Do communication & consensus make a
difference in performance?
• Only through their influence on
coordination
• What is associated with improvements in
coordination?
– History
– Evenness of communication, but not amount
– Level of functional knowledge
– Shared models of who knows what
Interactions
•
Shared models have greater benefits early
•
Even communication has greater benefits early
•
Models and communication volume substitute
Consensus X
Wave
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Early
Late
Questionnaire Wave
Low consensus
Hiigh
Communication X
Wave
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Early
Late
Questionnaire Wave
Uneven
communicaton
Even
Consensus X
Communication
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Low
High
Communication
Low consensus
High
Summary
• Communication and shared models influence
performance mainly through coordination
• Interesting results were the interactions predicting
coordination:
– Shared models and communication substitute. Using both
may hurt performance since too much effort is going into
communication.
– Shared models about the internals of the team and within-
group communication are important early in the team’s
history
– We’d expect team functioning will become routinized with
time
– Models about and communication with the external world
will probably become more important after routinization
• Shared models have unexplained, direct negative
association with declines in stock price
Limitations and plans
• Data problems
– Measures of accuracy, consensus, and dispersion were
constrained by low variation within a team
– Common input to measures of level and evenness
• Weak causal claims
– Panel design examined changes in process and outcomes, but
grossness of time granularity limits causal ordering
– Currently running lab experiments to manipulate
communication and recursors of shared models
• Scope
– Examined only small teams whose members had history
– Examined only shared models of who knows what
• Current work
– Current data collection is examining models of internal (e.g.,
who knows what) and external (e.g., competitive enviroment).
– Current data collection adds objective accuracy measures