A talk given at the SKIN3 workshop in Budapest, May 2014 (http://cress.soc.surrey.ac.uk/SKIN/events/third-skin-workshop)
Innovation or other policy-orientated research has tended to take one of two strategies: (a) work with high-level abstractions of macro-level variables or (b) focus on micro-level aspects/areas with simpler mechanisms. Whilst (a) may provide some comfort in the form of forecasts, these are almost useless for policy since they can only be relied upon if nothing much has changed. Although approach (b) may produce some interesting studies which show how complex even small aspects of the involved processes are, with maybe interesting emergent effects, it provides only a small part of the overall picture and little to guide decision making.
Rather, I (with others) suggest a different approach. Instead of aiming to produce some kind of "adequate" theory (usually in the form of a model along with its interpretation), that instead we aim at integrating different kinds of evidence and find the best ways to present these to policy makers in order to help policy-makers 'drive' by providing views of what is happening. Thus (1) utilising the greatest possible range of evidence and (2) providing rich, relevant but synthetic views of this evidence to the policy makers. Any projections should be 'possibilistic' rather than 'probabilistic' - showing the different ways in which social processes might unfold, and help inform the analysis of risks. The talk looks at some of the ways in which this might be done, to integrate micro-level narrative data, time-series data, survey data, network data, big data using a variety of techniques. In this view, models do not disappear, but rather have a different purpose and hence be developed and checked differently.
This shift will involve a change in attitude and approach from both researchers and those in the policy world. Researchers will have to give up the playing for general or abstract theory, satisfying themselves with more gentle and incremental abstraction, whilst also accepting and working with a greater variety of kinds of evidence. They will also have to stop 'conning' the policy world with forecasts, and refuse to provide these as more dangerous than helpful. The policy world will have to stop looking for a magic 'crutch' that will reduce uncertainty (or provide justification for chosen policies) and move towards greater openness with both data and models.
Towards Integrating Everything (well at least: ABM, data-mining, qual&quant data, networks and complexity science)
1. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 1
Towards Integrating Everything…
Bruce Edmonds
Centre for Policy Modelling
Manchester Metropolitan University
2. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 2
…well at least: ABM, data-mining, qualitative &
quantitative data, networks and complexity
science
3. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 3
Complexity Science and Policy Do
Not Mix Well..
Because (among other reasons):
• The phenomena of interest to policy makers is
complicated, complex and changing all the time and
thus very hard to “understand”
• Complexity Scientists and Policy People have very
different goals (understanding vs. action)
• They have to work in very different ways with respect
to very different sub-cultures/peers
• They are (mostly) unwilling to understand each
others’ world
• There is a tussle between them in terms of status and
power (control)
4. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 4
Some example problems
• “You have 3 months to give me your best forecast,
however preliminary”
• Policy makers are unwilling to outsource any control
over the process to researchers
• Important data is not available to researchers
• Policy makers already know what they will do, they
are just looking for a justification or story
• Complex models make policy debate difficult
• The outsourcing of blame: “The decision was made
based on the best scientific advice”
• The worry of researchers that the caveats underlying
their models will be lost/ignored
5. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 5
This talk is about some different
strategies for dealing with this
• That is, to step back, and look at what we are doing in
complexity science and how this might (or might not) ‘fit’
into the world of policy making and hence usefully inform
policy
• In other words, can our models and understanding be
useful to society and, if so, how
• Now, unfortunately we have only a little “data” about this
relationship – there has been relatively little observational
work on this
• Thus I will proceed by looking at some of the issues and
discussing possibilities only
• This is a synthetic talk, amplifying some trends and issues
already raised at this workshop then extrapolated to a
conclusion (that not all will be happy with)
6. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 6
Strategy 0: A natural language
discourse
• Issue is discussed in rich, meaningful language
• Any models are implicit and/or analogical
• Meaning of terms is often vague or ambiguous
– Advantages in terms of building consensus
– Disadvantages when working out what went wrong
• Debate is accessible to everyone
• But lack of cumulative knowledge development
• Depends upon “conceptual framework” and hence
can be influenced by fashion
• Not much good for anticipating radical change
• Good for integrating diverse evidence albeit informally
• But poor at dealing with complication/complex detail
7. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 7
Strategy 1: “predictive” “black-box”
macro-level models
• Model the system relating macro-level properties
of the whole system (using differential equations,
rate equations, systems dynamics etc.)
• With a view to predicting the effects of different
policy options (albeit with large error bounds)
• It is possible to make such models, of social
phenomena but understanding is often not the
main way of doing this but trial&error – in other
words model adaption (Nate Silver)
• But this only works if nothing essential changes –
these models only give “surprise free” projections
8. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 8
Strategy 2: partial, complex, micro-
level models used as an analogy
• Model some processes/aspects at the micro-level
system to observe the emergence of outcomes
• Does not easily relate to data, does not predict, and
remains quite abstract from the observed
• Can be used to understand/explore the possible
model “trajectories” of outcomes
• But the model remains more of an analogy, because
the mapping to any observed case is unclear and
remade by each interpreter
• This gives a “story” why, but is difficult to relate to
particular policy options/questions
• Tends to give “negative” conclusions w.r.t. policy
9. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 9
Kinds of Laws
• “How the Laws of Physics Lie” N. Cartwright,
1980
• Strategy 1 corresponds to “Phenomenological
Laws”, they match/predict the observed data but
do not explain
• Strategy 2 corresponds to “Explanatory Laws”,
they explain why things occur but do not predict
• You need both kinds of model but crucially also
the “bridging rules” learnt via acculturation
connect the two – this is often not explicit
• In a mature science formal derivations/theories
can be made between the two kinds of model
10. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 10
An illustration: Modelling a gas
T
V
P µT
V
The gas laws:
A Phenomenological Model
Picture of randomly moving, weakly
interacting elastic spheres
An Explanatory Model
Bridging rules
11. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 11
Back to Complex Social Science
• People are applying these two types separately,
indeed they are often in different academic fields
• The bridging rules have not been developed
• Broadly the policy world wants the properties of
strategy 1 models
• …whilst complexity science likes strategy 2
• This reflects the difference in their goals and both of
their wishes to retain control and avoid blame
• But neither, on their own, will satisfy the joint needs of
complexity and policy worlds, they are either: unable
to deal with structural change or they only provide a
kind of analogy as to what might happen
12. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 12
Robots in Uncertain Environments
• AI/robotics in the 60s/70s applied an approach
where the entity kept a model of its world and
tried to evaluate different alternative actions via
predicting their effects using the model
• However these were not good at coping with
unknown and uncertain environments
• Rather what turned out to work much better was:
– Using the world as a model of itself – frequent sampling
– Fast adaption in response to immediate conditions
– Different levels of abstraction and/or control
(subsumption architecture)
13. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 13
Strategy 3: “consistent” views of
what is happening
• Not general “theories” of social phenomena
• Rather well-understood and “consistent”
abstraction from the data that gives a “view” that
is designed for a particular purpose
• Behind this there will be a (maybe implicit) model
which explains how the resulting view is
generated from the source data
• But this model does not have to capture the
processes within the observed phenomenon
• And, crucially, it is not a projection forward in time
but a view of the current situation
14. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 14
This can involve different kinds of
data manipulation
Including:
• abstraction:
– losing information to make patterns/trends more visible
– selecting aspects relevant to the present purpose
• synthesis: integrating data from different sources
to produce:
– a more complete picture
– a more meaningful picture
• re-expression: manipulating the data to:
– suit the interpretative abilities of humans
– show how key aspects/properties relate/change
15. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 15
Desiderata for such “views”
• Timely: to be able to be produced reasonably
quickly (once designed!), so it is practical to use
• Adequate: to give enough information about the
current state and (maybe) how it is changing
• Consistent: in the sense that the same kind of
situation is reliably recognisable from its
presentation (given its purpose)
• Transparent: The meaning of what they show
needs to be clear enough to be learnable
• Well-founded: in the sense that it is clear how to
adapt it to new/changed data
16. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 16
Use of these “views”
• The algorithm for constructing these views might well
be derived from the results of extensive analysis,
simulation modelling etc.
• However, they are themselves a more straight-
forward representation of the data with a clear and
comprehensible account of their nature
• They are used to help “steer” policy by giving a richer
but timely sense of what is happening
• Policy makers (maybe initially with advice from
researchers) learn how to interpret the views
• They “own” the view machinery and its use, not the
researchers
• The views might be publically available, so that their
import may be discussed
17. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 17
An illustration of Strategy 3 in action
Modelling
micro-
aspects
Data
analysis
Expert
opinion
ABM and
other
analysis
Understanding
processual
possibilities
Measures,
tools and
visualisations
Policy
Decisions
Consequences
Wider Public
Policy WorldResearch World
18. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 18
In order to realise this strategy…
…we need to:
• integrate the streams of evidence/data
• discover what the possibilities in terms of social
processes are (using data mining, simulation
modelling etc.)
• use this to focus on what would indicate the early
onset of these processes occurring and their progress
• present tools for these using a variety of means
(visualisations, statistics, graphs, interactives etc.)
• but in a targeted and relevant manner
• then be willing to let go an others develop their use
ABM and other techniques
What is delivered to the policy world
19. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 19
Example: Understanding Innovation
Innovation is a multi-level phenomena, including:
• Individuals creating new things/ideas/innovations
• The spread of these over networks
• How innovations are received – how they ‘fit’ into the
conceptual and pragmatic frames of the potential
recipients
• How products/ideas combine to deliver services
• How some innovations are catalytic/tools – facilitate
the creation of many more innovations
• How innovations can change the affordances and
habits of a whole society and hence influence its
culture
• etc. etc.
20. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 20
Narrative Data
• Observational or personal accounts of what people do and their
motivations are valuable source material, especially for the
specification of the micro-level (agent behaviour)
• There is a wealth of such evidence within qualitative research,
albeit often “smothered” within (what seem to us as outsiders)
obscure jargon, issues, debates and pretentions
• This can be used to suggest some of the menu of strategies
people use and when they use them
• Are often relatively mundane and context-specific
• They might be mistaken, but are often a much better starting
point than the theories of academics
• Methods for using such narrative data to inform the specification
of agents are currently being worked upon
• These can, to some extent, bring some of the “mess” of
observed social life into ABM models
21. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 21
Psychological Evidence
• Often the constructs we use to make our agents
are ad hoc, coming from a “folk psychology”
• They are not checked for consistency against
what is known in psychology
• Unfortunately (for us) the psychological literature
is concerned with its own issues and debates and
does not often give the answers we need
• And it tends to look at different aspects in isolation
and gives no clue as to how they relate/integrate
• However it does provide weak constraints as to
the specification of agent behaviour and a
language within which to critique suggestions
22. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 22
big Data
• can provide fine-grained but noisy and incomplete
information about actions
• data-mining these can provide insights about how
people behave
• but provide richer insights when analysed with, or
compared against, other evidence and
hypotheses
• in particular, clustering of such data might indicate
some of the different kinds of behaviour
• each of which might be then abstracted to
suggest some different behavioural rules
23. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 23
Survey Data
• Existing surveys are often not very informative or
useful, only providing veiled hints for what people
might be doing/thinking
• The detail data from the survey is more useful
than the particular summaries presented in paper
• However, once the relevant behavioural
dimensions and strategies are known directed
surveys can be useful to determine proportion of
occurrence and conditions of occurrence
24. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 24
Network Data
• Direct network data is only sparsely available and
then often very incomplete
• Or derived from big Data where the connection
between links suggested by that data and social
connections in the target phenomena is weak
• But can be a useful validation check on the networks
that emerge from ABMs
• The relevant aspects of these networks could be
compared with those in available data
• However, comparing networks is non-trivial
• Even when sampled in odd ways, the sampling can
be mimicked within the ABM then compared
25. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 25
Participatory Validation
• When results are presented/developed with
stakeholders and/or subjects
• Can provide valuable input as to the meso-level
output/behaviours
• Context is important to provided –histories and
situations wherein the behaviour might occur so
that participants can give meaningful responses
• Often useful in getting indications of where a
model might be wrong
• Animations/visualisations/graphs are important to
give a fuller picture to participants
26. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 26
Time Series Data
• Are usually at an aggregate or group level
• Gives an idea of the dynamics one is looking for
• Are often quite derived/abstract
Aggregate Statistics
• Aggregate summaries of measures
• Give different projections of the data
• Are most useful when most closely linked to policy
inputs or (un)desirable outcomes
27. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 27
Integration via ABMs to “Views”
Micro-level
Narrative data
Psychology
Data-mining
Survey data
Network data
Participatory
input
Meso-level
Macro-level
Time-series
Aggregate
Statistics
Survey
summaries
ABM etc.
Archetypal
Stories of
Individuals
Complex
Visualisations
Key Global
Indicators
DeliveredtoPolicyWorld
Scenarios
28. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 28
Side Note: Social Context
• An obsession of mine
• Word “Context” is problematic, but roughly the
perceived kind of situation (e.g. keynote lecture)
• Context is not necessarily accessible to conscious
thought, reifiable, consistent or abstractable
• But some are socially entrenched and obvious
• People behave differently in different contexts!
• Identifying social contexts (via data mining,
observation etc.) is key to integrating many data,
particularly qualitative or specific data with more
generic/abstract patterns
29. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 29
Being A Modeller
• Status is largely accorded to higher end
abstractions rather than the concrete/mundane
A more balanced
science
Social
Simulation/Complex
ity Science
Abstract constructions
Concrete constructions
• We like to maintain control over our own
constructions including their subsequent use
• We inevitably see the world through our models
and so distort as we make progress
30. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 30
Some things I am suggesting to avoid
• Giving policy makers/advisers predictions is as
unwise as giving a sharp knife to a child – at best it
will not be useful to them and at worst it could cause a
horrible accident and it will be your fault
• Just because a model provides a good way of
thinking about things does not make it true.
• Policy makers expect scientists to only present only
(basically) correct models to them – not unvalidated
speculations or computational analogies (ways of
thinking about something).
• Even though you find a model really interesting and
revealing, do not attempt to share the model with
those involved in policy – they have fundamentally
different concerns and goals to you. This is not just a
matter of expressing things clearly/simply enough.
31. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 31
Conclusions
• The product should not be a model so much as a
representation of what is happening
• That is clear and the result of informed research
• But then ceded into the policy
development/assessment cycle, losing control
over it
• Thus to avoid relieving policy makers of the
burden decision making and future uncertainty
• In particular not to predict, even with error bars
• but to inform them better as to what is happening
• this will involve us shifting to data integration,
32. Towards Integrating Everything, Bruce Edmonds, Joining Complexity Science And Social Simulation For Policy, Budapest, May 2014. slide 32
The End!
Bruce Edmonds:
http://bruce.edmonds.name
Centre for Policy Modelling:
http://cfpm.org
The slides will be/are uploaded at:
http://slideshare.net/BruceEdmonds
Hinweis der Redaktion
Erik’s talk
From experience and Break-out session in a Complexity Science and Policy Conference in london
van de leew
Bert Droste-Franke’s indicates some of these difficulties
Many of the talks here
Not the Nancy Cartwright that does the voice of Marge in the Simpsons
like economic indicators: GDP, science score card of Muhamed and Matthias