Using multiple interacting computational models and multiple network modalities to predict the

advertisement
Using multiple
interacting computational models and
multiple network modalities to predict the
performance of the firm
Geoffrey P. Morgan
Kathleen M. Carley
MURI N00014-08-1-1186
Center for Computational Analysis of
Social and Organizational Systems
http://www.casos.cs.cmu.edu/
Short Version: What We Did
• Really important agent-based organizational models have attempted
to predict organizational performance due to the structural
constraints of explicit ties
• Other really important agent-based organizational models have
attempted to predict organizational performance due to the impact
of socialization processes.
• This new agent-based model integrates
– Structural constraints via authority networks
– Organization socialization processes
– Individual level bias implemented in hiring decisions.
• Preliminary results indicate that the new model is able to account
for both structural constraints and socialization processes.
– Structural constraints were noisy (many distinct structures were permitted)
– Impact of structure vs socialization is not intended to be definitive.
March 2012
G.P. Morgan & K.M. Carley
2
Why is this important?
• Some work-oriented ties are easy to capture
– Org Charts, for example
• Other ties can be much harder to capture
– Examples
• “Hangs out at the same bars”
• “Goes to the same church”
• “Shops at the same stores”
– Yet these ties can also share information
– Unclear which ties actually have salience in any particular
environment
• A more holistic model, which includes easy-to-capture
structural ties as well as implications of harder-tocapture ties seems like an approachable middle-ground
March 2012
G.P. Morgan & K.M. Carley
3
Leveraged Models, 1 of 4
• The Mutual Learning Model (March, 1991)
– Influential Organizational Model: 8000+ Cites (Google Scholar,
March 2nd)
– Organizations are presented as performing better if the
organization’s knowledge is more accurate.
– The organization’s knowledge is not an aggregate of the
knowledge of its individuals, but instead is inferred from highperforming actors in an error-prone process.
– The model makes no claims about how the members of the
organization are organized.
Organization
?
Agents
March 2012
G.P. Morgan & K.M. Carley
4
Leveraged Models, 2 of 4
• The Hierarchical Garbage Can (Carley, 1986a)
– Extended Cohen, March, and Olsen’s intellective Garbage Can
Model (1972)
• Another influential Organizational model: 5000+ cites
• The Garbage Can Model got its name by suggesting that
organizations can be thought of as a garbage can fed by four
different streams as those products arrive. The streams were
problems, solutions, participants, and choice opportunities.
• They were interested in how the elements of these different
streams would glom together in “the can” even if they were not
perfect fits.
• Results very sensitive to the timing of the streams.
A hodepodge of stuff in this garbage can
How will it all stick together?
March 2012
G.P. Morgan & K.M. Carley
5
Leveraged Models, 3 of 4
• Hierarchical Garbage Can, incorporating theoretical work
by Padgett (1980) on how “the Garbage Can” would
behave with hierarchies, took a knowledge-oriented
perspective and examined the flow of ideas through
organizational hierarchies.
Hodgepodge
March 2012
Ideas
G.P. Morgan & K.M. Carley
Hierarchies as Graphs
6
Leveraged Models, 4 of 4
• The Participation Model (Morgan, Morgan, and Ritter,
2010) is a mathematical model to moderate individual
action.
• This model suggests that the larger context moderates
the actions of individuals and proposed key factors from
the sociological literature.
• The model matched historical data – decreased average
performance and increased variation (pattern validity)
To act or not to act?
…
It depends.
March 2012
(Is my side winning?)
G.P. Morgan & K.M. Carley
7
The New Model – The Unified
Hierarchical Model
• We took insights from each of these models and created a new
model.
• The new work in this model, in particular, is taking the “Org as
Agent” Mutual Learning Model and reconciling it with the “Org as
structured hierarchy of agents” from the Hierarchical Garbage Can.
More than the sum of its parts?
Or not?
March 2012
G.P. Morgan & K.M. Carley
8
Lite-Docking the Models
MLM
HGC
Par
Characteristic
Organization in an environment
Environment changes over time
Organization learns from agents
Organization socializes agents
Agents leave Org at random
Org replaces agents at random
Org has explicit authority ties
Team members generate info
Information travels along ties
Information transfer has error
Org removes under-performers
Org can have structural flaws
Explicit access constraints to info
Context moderates agent action
Dyad distance moderates action
Spatial distance moderates action
Social distance moderates action
Agents implement homophily bias
Committee makes hiring choice
Org accuracy measured over time
CEO accuracy measured over time
Structural flaws tracked over time
March 2012
MLM
♦
♦
♦
♦
♦
♦
HGC
Par
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
G.P. Morgan & K.M. Carley
UHC
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
♦
9
Model Initialization
• A group of agents is specified and assigned to one of
four organizational levels.
• Reports-To Relationships are determined per run when
the organization is created. Rules were loose enough
that many structures were permissible.
March 2012
G.P. Morgan & K.M. Carley
10
Model Operation: 2 Learning
Mechanisms
HGC
Hierarchical
Transmission
Organizational
Inference
MLM
Org more likely to change as
more High Performers disagree with it.
HP: Disagree
HP: Agree
Org Members
BitBit
B: A:
Will
Change
not change
Likely
March 2012
G.P. Morgan & K.M. Carley
11
Model Operation: Bias in Hiring
“Don’t
kindcall
of office
us, we’ll
would
call
youyou.”
like?”
Yes “What
No
Yes
No
No
Yes
Prospective Team Members
March 2012
G.P. Morgan & K.M. Carley
12
Virtual Experiment
Variable
HGC
Values
Structure ACME: m = 10, g = 2;
ZENO: m = 6, g = 6
Socialization (o) 0, .05, .9
MLM
Constants
Complexity (c)
Redundancy (r)
Simulation Length (s)
Grace Period (l)
Turbulence (t)
Perception Acc (p)
Update Capacity (u)
Staff Agreement (a)
March 2012
50
.3
100
5
.05
.9
3
.5
Total Combinations
G.P. Morgan & K.M. Carley
#
2
3
1
1
1
1
1
1
1
1
6
13
Results
March 2012
G.P. Morgan & K.M. Carley
14
Socialization has a U-Shaped
Curve?
Copyright owned by
Academy of Management Journal
March 2012
G.P. Morgan & K.M. Carley
15
Summary
• Our model attempts to integrate both structural
constraints and socialization processes into a single
model while also including a bias mechanism for new
hires.
• Our results indicate that the new model is able to
account for both structural constraints and socialization
processes.
– Structural constraints allowed for significant noise.
– Impact of structure vs socialization is not intended to be
definitive.
– Performance peaks because only team members are replaced,
later versions will allow all staff members to leave.
March 2012
G.P. Morgan & K.M. Carley
16
Download