2. When ABM are worth it
ABM most appropriate for systems with:
Interactions (between agents)
Heterogeneity (in agents’ context)
Organized Complexity (i.e., middle-numbered)
Argument
these properties mean a narratives can
explain how ABM structure produces
emergent system-level patterns
http://landscapemodelling.net
3. ‘Generative’ Simulation
Specification of micro-level properties, or
rules of element interactions, used to
generate observed macro-level patterns
“If you didn’t grow it,
you didn’t explain its emergence”
Epstein (1999, p.43)
http://landscapemodelling.net
4. Statistical Portraits
Artificial Anasazi
Axtell et al. (2002)
Statistical Portraits
of Pattern
http://landscapemodelling.net
5. Elaborate Black-Boxes?
“There are some warning signs here in the ABM
enterprise insofar as that greatest criterion of
‘success’ – and the claim to novelty itself – is that
patterns are produced as outcomes whereas the
intermediate process (i.e. interactions between
simple rules) which leads to structure is shrouded.”
(Clifford 2008, p. 682).
http://landscapemodelling.net
6. ABM are Event-Driven
Event: any interaction between modelled
entities that results in a change in state of
at least one entity attribute
Direction
Stress Level
of travel
Location
Wealth Any other
attribute
http://landscapemodelling.net
7. ABM are Event-Driven
Event: any interaction between modelled
entities that results in a change in state
Events are consequences of code
executed in context
http://landscapemodelling.net
8. ABM are Event-Driven
Event: any interaction between modelled
entities that results in a change in state
Events are consequences of code
executed in context
Event-driven: sequences of low-level
events produce system-level patterns
http://landscapemodelling.net
9. Narrative Explanation
Explain causes of events from numerous,
and potentially distal sources through a
coherent sequence of prior events (Cleland 2011)
Historical Natural Science distinguished
from ‘Classical’ Experimental Science
Narrative shows how a focal event or state
came to occur by fitting it into a coherent
account of a sequence of preceding events
http://landscapemodelling.net
10. What is a narrative?
Narrative
Understanding
Events
http://landscapemodelling.net
11. What is a narrative?
Narrative
…may move back and forth between
accounts of low-level events and
system level (statistical) summaries to
show how they are linked
… is not simply a chronicle of events
http://landscapemodelling.net
12. An example
Breeding synchrony in bird colonies
Jovanni and Grimm (2008) Proc. R. Soc. B
http://landscapemodelling.net
13. An example
Breeding synchrony in bird colonies
Jovanni and Grimm (2008) Proc. R. Soc. B
Hypothesis:
interactions between neighbouring
birds’ stress levels drives synchrony
http://landscapemodelling.net
17. Heterogeneity in context
All parameters apply to all birds identically
Only difference is initial stress level
Narratives more useful with heterogeneity
Heterogeneity varies context of interactions
Consequently, events are more important
Modify model
so birds arrive at colony at different times
different neighbours influence stress level
http://landscapemodelling.net
23. Potential Issues
(Re)Introducing uncertainty?
From formal model to informal language
Which narrative do we choose?
How do we know if our narrative is ‘good’
(enough)?
Loss of objectivity?
Highlights subjectivities of modelling
But maybe this is a good thing…
http://landscapemodelling.net
24. Summary
Why simulate individuals and then report
aggregated patterns alone?
Spatially-explicit model without maps
Explaining ABM events via narrative can
reveal process
Millington et al. (2012) Geoforum
james.millington@kcl.ac.uk
http://landscapemodelling.net