The Role of Agent Based Modelling in Facilitating Well-being Research: An Introduction and Existence Proof

http://www.simian.ac.uk
Edmund Chattoe-Brown (ecb18@le.ac.uk)
Department of Sociology, University of Leicester
The Role of Agent Based Modelling in
Facilitating Wellbeing Research: An
Introduction and Existence Proof

2 http://www.simian.ac.uk
Thanks
• This research funded by the Economic and Social
Research Council as part of the National Centre
for Research Methods (http://www.ncrm.ac.uk).
• To my colleagues in the Wellbeing and Health
Research Group, Department of Sociology,
University of Leicester, for comments on a
previous version.

Problems with interdisciplinarity
• Disagreement on what is important or interesting. Here we seem
to be agreed.
• Disagreement on what counts as an “acceptable” theory (Rational
Choice versus Marxism or Post Modernism). Here I can only state
a position and say I am a “middle range” empiricist.
• Disagreement on what counts as an acceptable research method.
Here is where my paper “is at”.
• We cannot seem to resolve these debates within the existing
framework. (I may be able to show why.)
• Important problems are partitioned arbitrarily across the social
sciences by these disagreements to no obvious advantage.

A two part solution 1
• Rejecting theory (old style). Decision making example.
• The difference between decision making (collect or
acquire information, process it, decide and act) and
Rational Choice Theory is assumptions not evidence.
• Compare what a sociologist might reasonably say
about network flows of information/influence or a
psychologist about rationalisation and weakness of will.
• But this iconoclasm is no use without a technique that
allows real theory (new style) to be built.

A two part solution 2: Schelling
• A technique known as Agent Based Modelling (ABM). Social theories as (a particular
kind of) computer programmes rather than equations or narratives.
• I’m not presenting this because it represents social reality but because it is easy to
explain and gives a good point of access to important issues.
• “Agents” live on a square grid so each has maximum 8 neighbours.
• There are two “types” of agents (pink and white) and some grid spaces are vacant.
Initially agents/vacancies are distributed randomly.
• All agents decide what to do in the same very simple way.
• Each agent has a preferred proportion (PP) of neighbours of its own kind (0.5 PP
means you want at least half your neighbours to be your own kind - but you would
accept all of them i. e. PP is minimum.) Vacant grid spaces “don’t count” which is why
the PP is a fraction not a number.
• If an agent is in a position that satisfies its PP then it does nothing otherwise it moves to
a vacancy chosen at random.

Initial random state

Clustering
Schelling (early
seventies) was
interested in urban
residential
segregation of
ethnic groups in
the US.

Surprising results?
• What is the smallest PP that will produce clusters?
• What happens when the PP is 1?
• Even in such a very simple system is is very hard to
“gross up” from individual behaviour in context or “drill
down” from aggregate patterns.
• This is why statistics and ethnography argue past each
other with such futility.
• Most real social systems are non-linear/emergent and the
levels of description are thus isolated by the limitations of
our “common sense” reasoning in crossing the gap.

Simple individuals/complex system
Individual Desires and Collective Outcomes
-20
0
20
40
60
80
100
120
0 50 100 150
% Similar Wanted (Individual)
% Similar Achieved (Social)
% similar
% unhappy
Counter-intuitive
macro (social)
results from
simple micro
interactions. A
non-linear (and
complex)
system.

Why?
• Because PP is a minimum, people are always happy
“inside” a cluster of their own kind.
• If a cluster is “full” (no internal vacancies) then it cannot
be “disrupted” except at the edges.
• Whether clusters form/persist thus depends on whether
their shape is compatible with the PP for “edge agents”.
(No “sharp corners”: Defines minimum cluster size?)
• When PP is 1, no shape of the cluster edge is
compatible with the satisfaction of edge agents so the
cluster cannot maintain itself. (PP heterogeneity?)
• An aggregate entity (the cluster) thus becomes a
organising principle for individual behaviour: This is a
very concrete example of what sociologists call (often
very verbosely) “structuration”.

An aside on data
• If you did want to calibrate and validate this model
you would need both:
• Micro data on decision making, environment
perception, psychological constructs on
contentment (experiments, observation,
interviews).
• Macro data on clustering (surveys, spatial
statistics).

Welcome to WellbeingWorld

How WW works
• We do not try to explain the arrival of money and goods
(“stuff” and “nonsense”). These drop from the sky.
• Agents go shopping and buy/consume according to
“rules of thumb”. (“When you can buy x units”. “In each
period, consume y units”.)
• Nonsense gives much more utility than stuff per unit.
• The most basic case of all is where agents adapt their
strategies by comparing a random strategy mutation on
“historical” wellbeing (individual learning).

More on WW
• Agents are not “told” that nonsense is “better value”
than stuff. They infer this by experience.
• At the same time, without being told, they learn to
spend all their money. This is like real life which must
be learned from “inside” while lots is going on.
• This is an absolutely minimal model in many respects:
There is no price competition, no diminishing returns,
no “non separable” consumption (you can’t shop at
Tesco out of town without a car) and so on.

Beginning synthesis
• Individual learning (psychology, economics): Basic model.
• Social learning (agents exchange “candidate” rules of thumb at interaction:
Sociology, social psychology).
• “Acts of God” (external environment): White blocks “hit” agents with good
or bad outcomes. (Coming out of the house to find out someone has
ripped off your wing mirror. Winning the pools.)
• Social interaction (sociology, psychology): Agents can directly “transmit”
wellbeing to each other.
• Cognitive process (psychology): wellbeing “decays” endogenously.
• Others? But based on evidence please. (See last section.)
• Interesting general question: To what extent can people accurately
attribute wellbeing to its “actual” causes? (Children who cannot enjoy
anything when hungry or tired.)

Some experiments: Background
• Nice to be able to “run the world” as often as you
want.
• Comparing total wellbeing (all agents, all periods)
after 5000 periods (area under curve) as measure of
adaptation effectiveness.
• Five typical runs: 24,480,244, 21,218,955, 26,951,127,
32,674,847, 22,486,960. Significant variation but not
totally path dependent.

Some experiments
Effects of Different Learning Mechanisms
and Wellbeing "Processes"
0
5,000,000
10,000,000
15,000,000
20,000,000
25,000,000
30,000,000
1
Mechanism Combinations
Total Wellbeing (Arbitrary
Units)
Ind
Ind/Im
Ind/Ext
Ind/Im/Ext
Ind/Int
Ind/Im/Int
Ind/Int/Ext
IIIE

Some experiments: Discussions
• More “difficult” or misleading worlds (where you may
“misattribute” poor wellbeing from “bad news” to inept
consumption) are harder to learn but they are learnt.
• Sometimes social learning is helpful (external events),
sometimes it is a hindrance (simple world) and sometimes
it doesn’t seem to make much difference (world with
wellbeing through interaction and that combined with
external events). No joy for economics but not unlimited joy
for sociology and psychology either!
• Note we are exploring the interplay of dynamic processes
not static factors: Is this qualitative or quantitative?

Other useful outputs
Sensitivity analysis: We can
get some sense of how
much it matters that a
particular variable value or
process specification was
more or less a guess.
Clear evidence ex post that this is
a complex (non-linear) system in
which scaling is non trivial and
common statistical assumptions
may not work.
Effect of Memory on Wellbeing
0
1,000,000
2,000,000
3,000,000
4,000,000
5,000,000
6,000,000
7,000,000
8,000,000
9,000,000
1
Memory Length
Total Wellbeing
Mem 5
Mem 10
Mem 20
Effect of "Wane" on Wellbeing
346,513,929
53,243,930
25,562,427
6,752,179
0
50,000,000
100,000,000
150,000,000
200,000,000
250,000,000
300,000,000
350,000,000
400,000,000
1
Wane Multiplier
Total Wellbeing
Wane 1
Wane 2
Wane 3
Wane 6
Effect of Income on Wellbeing
10,740,317
101,474,423
843,388,000
1,462,739,7
29
0
200,000,000
400,000,000
600,000,000
800,000,000
1,000,000,000
1,200,000,000
1,400,000,000
1,600,000,000
1
Income Multipliers
Total Wellbeing
0.5x
1x
2x
3x
4x

What about real data?
Just a few examples: Distributions from large scale surveys,
rules of thumb ethnographically, biographical data from self-
report diaries.
A simulation that fits one or two of these is probably just lucky
but one that fits all of them is probably “onto something”.

Methodology (G+T)

Caveats
• This model was developed quickly to show what
might be done using “social science common sense”.
I don’t profess to be even an amateur in wellbeing
research at thus stage.
• Thus it is still a “toy” model. Not only is not calibrated
or validated, it is not even based on a systematic
review of what is known.
• The paper is an existence proof because it shows
what might be done and perhaps why it is worth
doing it.

Obvious over simplifications
• Real social structure (not just social networks but
real places and groupings like “workers at Leicester
University”).
• “Closing” the model as appropriate. Money doesn’t
drop from the sky but comes from work which
shapes wellbeing directly as part of social structure.
• Interplay between kinds of consumption
(requirements, large scale plans and saving,
maintenance and “spoiling”.)

How might research proceed?
• Version 0 model which synthesises all “big” wellbeing processes
as identified as the “peaks” in the literature.
• Attempt to narrow range of parameter values from existing
literature (including proxies: close friends example).
• Experimentation to find out which behaviours might be tested
and which data absences are most harmful.
• Model informed data gathering (needle holding example).
Interesting question: Is this a new kind of ethnography?
• Calibrate and validate: This can “work”.
• Repeat if necessary?

Conclusions
• ABM can start to bridge the arbitrary gaps between disciplines in
terms of interests and acceptable “theories” and data.
• Many topics are too important for this not to be done.
• Building an ABM is heuristically fertile as it poses “rigorous”
challenges to preconceptions built into existing theories (level of
self awareness example).
• In principle, the ABM methodology offers falsification
(unfashionable but still valuable).
• Institutionally, it may provide a framework which allows different
disciplines and methodologies to work together through relatively
“neutral” process specifications.

Further resources
• Simulation Innovation, A Node (Part of NCRM: research,
training and advice): <http://www.simian.ac.uk>.
• NetLogo (software used here, free, works on
Mac/PC/Unix, with a nice library of examples):
<http://ccl.northwestern.edu/netlogo/>.
• Simulation for the Social Scientist, 2nd edition, 2005,
Gilbert/Troitzsch. [Don’t get first edition, not in NL!]
• Agent-Based Models, 2007, Gilbert.
• Journal of Artificial Societies and Social Simulation
(JASSS): <http://jasss.soc.surrey.ac.uk/JASSS.html>.
[Free online and peer reviewed.]
• simsoc (email discussion group for the social simulation
community): <https://www.jiscmail.ac.uk/cgi-
bin/webadmin?A0=SIMSOC>.

The Role of Agent Based Modelling in Facilitating Well-being Research: An Introduction and Existence Proof

More Related Content

What's hot (14)

Similar to The Role of Agent Based Modelling in Facilitating Well-being Research: An Introduction and Existence Proof (20)

More from Edmund Chattoe-Brown (20)

Recently uploaded (20)

The Role of Agent Based Modelling in Facilitating Well-being Research: An Introduction and Existence Proof