Putting people into models
Starting with qualitative models
Ingrid van Putten
CSIRO – Marine and Atmospheric research
(Hobart- Australia)
CSIRO Mathematics, Informatics, and Statistics
One of a number of modeling approaches
If you use it depends on
What can different types provide?
(Generality, Precision, Realism)
What you want from model?
(Understand, Predict, Modify)
Don’t need much data
QUALITATIVE MODELS
Good for combining biophysical and human domain –
but philosophically – can we
actually model humans?
PRECISION
GENERALITY
REALISM
Richard Levins
1966
PRECISION
GENERALITY
REALISM
STATISTICAL MODELS
PRECISION
GENERALITY
REALISM
MECHANISTIC MODELS
Philosophical perspective
Can we model human behaviour?
Behaviorism: probably…?
Ivan Pavlov
(1849 –1936)
Burrhus Frederic Skinner
(1904 – 1990)
Behaviour shaped by response to environmental stimuli
Human beings perceive, assess, decide, and act.
Modellers need algorithms for each stage
But how do we “observe” and
interpret what human beings do
Metaphysics: never…!
Aristotle
Plato
Human beings aren’t reducible to any description.
Transcendental nature of ‘self’ and cognition.
Assuming we can model human behaviour
How do we observed it?
Inductive reasoning
Based on observation
Inference of general principles
or rules from specific facts
Deductive reasoning
Based on interpretation
Inference of specific facts from
general principles or rules
What’s
next?
“Causal black box”
“Cognitive
white box”
Do we need to know what goes
on
in the cognitive box (the brain) when
Formal logic
Empirical heuristicsmodelling people?
compliant agents
based agents
Do we need to know what goes on in the cognitive box
when modeling the way people make decisions?
Ask an economist ……
Faced with a
problem
Gather all information
necessary for rational
judgement
Uncertainty
Make decision
Person acts rationally in complete
knowledge out of self-interest and
the desire for wealth
homo economicus
Not much gain from knowing what goes on in the cognitive box
Psychologists say we do need to know
about the cognitive box
When people are faced with a complicated judgment or
decision, they often simplify the task by relying on
heuristics, or general rules of thumb (shortcuts)
Amos Tversky and Daniel Kahneman (1972)
Gather all information
necessary for rational
judgement
Uncertainty
Heuristic
(shortcut)
Make decision
The rules explain how people make decisions, come to judgments, and solve problems
The rules can be learned or hard-coded by evolutionary processes.
Cross fertilization between economics and psychology
Behavioural economics
Study the effects of social, cognitive, and
emotional factors on economic decisions and
resource allocation
Concerned with the bounds of rationality of the economic agents
In some situations, heuristics lead to predictable
biases and inconsistencies
Gather all information
necessary for rational
judgment
Uncertainty
Heuristic
(shortcut)
BIAS
Make decision
In other words ……
Behavioural rules in psychology work well under most circumstances,
but in certain cases lead to systematic errors or cognitive biases
Some examples of cognitive biases
Decision-making and behavioural biases
Loss aversion (endowment effect) – people demand much more to
give up an object than they would be willing to pay to acquire it
losing $100 affects your level of happiness much more than winning $100
Probability and belief biases
Outcome bias – People overestimate small probabilities and
underestimate large probabilities
Low frequency events (such as smallpox, poisoning, and botulism) are
overestimated (by a factor of 10), while high frequency events (such as
stomach cancer, stroke, and heart disease) are underestimated
Social biases
False consensus bias – People tend to overestimate the degree to
which others agree with them
(Lichtenstein et al. 1978
Memory biases
Consistency bias – people often incorrectly think past attitudes and behavior
resemble present attitudes and behavior.
How do we know cognitive biases happen?
Do experiments with people to find out how
they might behave in different situations
Example of an experiment to establish cognitive bias
Anchoring – the tendency of people to rely too heavily, or "anchor,"
on one trait or piece of information when making decisions
Question
Guess the percentage of African nations that are
members of the United Nations
Group 1
Was it more or less than 10%
25% on average
Group 2
Was it more or less than 65%
45% on average
Before the experiment Write down the last two digits of your social security number
Consider whether you would pay this number of dollars for items value
(e.g. wine, chocolate, computer equipment) with an unknown
People with higher
Group 1
60 to 120% higher payment
numbers (e.g. 85)
offered for the goods by people
People with lower
with higher numbers
Group 2
numbers (e.g 20)
Question
Why do we care about cognitive biases?
Raghu mentioned it – for instance climate change communication
Things like confirmation bias which describes how people
are more likely to search for or accept information that
supports pre-conceived beliefs.
Google search histories illustrated this:
Believers will tend to use search terms
“climate change proof”
disbelievers terms such as
“climate change myth”.
Both believers and disbelievers
are presented with search results
that support their original belief.
Not only do we look for information that confirms our
preconceived ideas but we also believe that everyone
else believes the same as us?
False-consensus bias
We overestimate the prevalence of our personal opinions in society while we
underestimate the prevalence of beliefs that conflict with our own
7% of Australians believe that climate change isn’t happening at all.
That same 7% believe that almost 48% of the population agree with them.
78% believe climate change is real.
- 63% believe that climate change is already happening;
- 15% believe that climate change will happen
in the next 30 years
15% are unsure if climate change is real
2011 paper on public perceptions of climate change by the CSIRO
Some of the biases Skeptics accuse Believers of
OVERCONFIDENCE - in the predictions of their computer models.
ILLUSION OF CONTROL - Believers think that human reductions of greenhouse
gases will make a large enough contribution to reduce global warming, but
Skeptics think that’s an illusion.
LOSS AVERSION - Skeptics claim Believers overestimate the costs of
warming (compared to the benefits).
BANDWAGON EFFECT the tendency of Believers to believe climate change is
happening because many other people believe the same.
AVAILABILITY BIAS - “because believers think of it, the believers think it must be
important."
CONFIRMATION BIAS- Believers search for or interpret information in a way that
confirms their preconceptions
Why is it useful to know about cognitive biases
As Raghu said – we can change peoples mental models - knowing
about (both sceptics and believers) cognitive bias will help
Why is it useful to know about mental shortcuts that psychologists
study (heuristics) when modelling human behaviour?
As Rashid said – economics can develop incentives to change behaviour knowing about mental shortcuts people take in making decisions will help
develop incentives that work
As Eileen said – we need coupled models to go into the future - knowing
as much realistic information about the way we make decisions will be
central to that
Qualitative modelling is one of a number of approaches to couple human
to bio-physical systems
- Not data hungry
- Intuitively simply
- can follow easily from conceptual modelling
- can be developed with the people represented in the model
Introduction to qualitative modeling
Systematically developed by Richard Levins (1966)
Qualitative models are based on signed digraphs
Sign Directed Graphs
(Signed Digraphs)
Predator-Prey
A few historically significant scientific discoveries had to happen before
qualitative modelling came along
Liber Abaci (Book of Calculation)
“A certain man put a pair of rabbits in a place
surrounded on all sides by a wall. How many
pairs of rabbits can be produced from that pair
in a year if it is supposed that every month each
pair begets a new pair which from the second
month on becomes productive?”
Leonardo Fibonacci in
1202 (age 32)
Fibonacci number sequence:
1 1 2 3 5 8 13 21 34 55 89 144
Geometric or Exponential Increase
Essay on the Principle of Population
Populations Increase Geometrically (e r t )
Resources Increase Arithmetically (x + y)
Thomas Malthus
In 1798 (age 32)
"The power of population is indefinitely
greater than the power in the earth to produce
subsistence for man"
Lotka-Volterra type equations describe the
Darwinian evolution of a population density
Charles Darwin
PREY
dN1
 N1 birth  1, 2 N 2 
dt
PREDATOR
dN 2
 N 2   2,1 N1  death 
dt
Predator-Prey
Alfred Lotka
1925
Vito Volterra
1926
Mathematical relationship
dN1
 N1 birth  1, 2 N 2 
dt
dN 2
 N 2   2,1 N1  death 
dt
Lotka and Volterra 1925-1926
Richard Levins
1966
Community Matrix
0
-α1,2
+α2,1 0
Levins 1968
Signed Digraph
-α1,2
+α2,1
Levins 1974
Qualitative modelling
Positive
effect
Negative
effect
Predator-Prey
Competition
Mutualism
Amensalism
Commensalism
Self-Effect
Change in
Community Matrix
1. Small fish
2. Large fish
3. Fishery
Due to interaction with
1
2
3
-a11
-a12
0
+a21
-a22
-a23
0
+a32
-a33
3
Fishery
2
Large fish
1
Small fish
Self effect
Change in
Community matrix - signs only
1. Small fish
Due to interaction with
1
2
3
0
2. Large fish
+
-
-
3. Fishery
0
+
Self effect
What can qualitative modelling tell you
– beside increases and decreases?
1
Qualitative models can identify key drivers of change and
predict the direction (+, - , 0) of response to change
Press perturbation: shift in parameter leading to new equilibrium
Pulse perturbation: shock to population or variable leading to transient
dynamics
2
Assess model stability (important for assessing the reliability of
predictions) – if strong positive feedback system then unstable
3
Qualitative modelling can be used to identify data gaps and
hypotheses for further investigation
Additional benefit of qualitative modelling
Qualitative models are relatively easy to produce with
stakeholders (next step to building a conceptual model)
“…a very underrated tool in biology and social science” (M.L. Cody 1985)
Marine sectors
Marine environment
(ecological groups)
Climate
drivers
Australian example of qualitative model
Connect climate change drivers, to marine environment and marine sectors
(‘expert model’)
Temperature
-
Wind
Cyclones
& storms
Other
industrial
use
Renewable
energy
Currents
+
+
Rainfall
Emergent
species
Sea level
rise
Pests &
diseases
Retained
species
Non-retained
species
+
Ecosystem
integrity
Commercial
fishing
Aquaculture
Charter
fishing
Marine
tourism
Recreational
fishing
Traditional
owners
non-fishing based
recreation
Marine sectors
Marine environment
(ecological groups)
Climate
drivers
Build same model with community members
Temperature
Wind
Currents
Rainfall
Emergent
species
Retained
species
Cyclones
& storms
Sea level
rise
Pests &
diseases
What did we learn?
Incomplete understanding of the whole system
Will help shape communication/education/information
Ecosystem
Non-retained
species
integrity
Commercial
fishing
Aquaculture
Charter
fishing
Marine
tourism
Recreational
fishing
Traditional
owners
Other
industrial
use
non-fishing based
recreation
Renewable
energy
 The pathway by which the fishers
thought climate change affected
them (fisher’s mental model)
Climate change
Sea
temperature
Currents
Retained
species
Emergent
species
Fish
abundance
Price of fish
Profitability
Commercial
fishing activity
 Climate is not only thing that drives
fishing activity (fisher’s mental model of
where it fits in)
Retirement
funding options/
alternatives
Vessel
Size
#1
Vessel
ownership
Family quota
ownership
Quota
ownership
Fixed cost
Harbour access
channel sand
build up
Sea
temperature
Family
Currents
Retained
species
fishing
Quota ownership
history
characteristics
Pass quota
down
Climate change
Bank lending
rules
Emergent
species
Quota trade
Method of
lease quota
characteristic
Admin.
monitoring
requirements
Fishing
pressure
trade
#2
Season
Price of
lease quota
Variable cost
Fish
abundance
Imports
Exchange
rate
Price of fish
Exploratory
licence rules
Public works
funding
Government
TAC levels
Govt dept
resources
Age
Oil & gas industry
development
Important to understand how these things fit together
Diversification
Alternative
income
Work
Exploratory
to
Profitability
options
if Access
we
want
to
use
policy
to
change
the
system
improve
it
–
earning
options
harbour
opportunities
fishing
or make it more robust
#4
Commercial
fishing activity
#3
Example of how Qualitative models can provide powerful insight
when you want to implement policy to improve the system
Benguela Ecosystem - effects of seal cull on hakes
(-)
Merluccius capensis
Live in shallow water
Hakes model
+
(-)
+
Mc
J
Juveniles
+
Mp
J
Juveniles
Mc
A
Adults
Mp
A
Adults
Merluccius paradoxus
Live in deep water
Yodzis 1998
Benguela Ecosystem - effects of seal cull on hakes
Merluccius capensis &
Hakes model
Merluccius paradoxus model
+
(-)
+
(-)
+
+
+ -
Shallow
Deep
Punt 1997
Another example of qualitative model in fisheries
How QMs can address hypotheses regarding reduced banana prawn catch
What happens when the model gets perturbed
•
•
•
•
•
•
Weipa region
Reduced banana prawn abundance from recruitment overfishing,
Reduced banana prawn abundance from change in environment,
Reduced banana prawn abundance from pollution.
Reduced fishing effort in Weipa.
Reduced catchability from prawns remaining inshore,
Reduced catchability from reduced aggregation or “balls”.
Example of qualitative model in fisheries
Banana Prawn Subsystem
PL
PA
Larvae
Adults
PJ
Juveniles
PJ
Pm
Predators
N3
Prawns
N2
Prawn food
N1
Maturing
PA
Example of qualitative model in fisheries
Pr
Pr
off-shore
PL
PA
Op
Pr
Nut
estuary
PJ
Pm
Nut
in-shore
Banana Prawn
biological
Subsystem
Example of qualitative model in fisheries
Eff
Rec
(est)
Rec
(oce)
CPUE
Com
$
catch
Human system
Pr
Pr
PL
Op
PA
Pr
Nut
estuary
PJ
off-shore
Pm
Nut
Tur
(est)
Man
hab
Biological
system
in-shore
Rain
Tur
E
(oce)
Rain
L
Environment
& habitat
Sal
CSIRO Mathematics, Informatics, and Statistics – Jeff Dambacher
Example of qualitative model in fisheries
E
Rec
(est)
Rec
(oce)
CPUE
Com
Commercial fishing
economic system
Recreational
fishing system
$
catch
Pr
Pr
+
PL
-
Op
PA
Pr
?
Nut
estuary
PJ
Pm
Nut
Tur
(est)
Man
hab
Biological
system
in-shore
Rain
Tur
E
(oce)
PERTUBATION
0
off-shore
Rain
L
Environment
& habitat
Sal
Why use qualitative modelling?
1
Few data required – only need signs of the interactions
Positive effect
Negative effect
Reciprocal effects
S. Metcalf, Murdoch University
Fish
stocking
Fish
populati
on
Stocking of rivers with
fish increases the
abundance of fish
Birth
rates
Female
education
Female education will
decrease birthrates
Policy
present
on
street
Cars
stolen
Police on the street will
decrease the number of cars
stole and if more cars get
stolen this will increase police
presence
Why use qualitative modelling?
2
Any type of interaction cay be included in qualitative models
(biological populations, whole ecosystems, groups of people,
economic variables, nutrients, social and demographic
characteristics).
Birth
rates
3
Female
education
Can investigate direct and indirect interactions and their
effects on the dynamics of the system
Indirect interaction
Direct
interaction
4
Wealth
Qualitative models are excellent for producing
with stakeholders (participatory modelling)
Why involve the community in the modeling exercise?
(Participatory modeling)
Stakeholders learn more about:
How to structure and formulate their ideas
Understand situation and possible options
How to understand, discuss and cooperate with others
Scientists learn more about:
Stakeholder’s views and social behavior
Ways of translating research into policy practice
Policy makers benefit as legitimacy of models is enhanced
Direct integration into the decision-making process
Social and scientific validation
Policy makers benefit from what the scientists and stakeholders have
learned by developing the model together and from the legitimacy
gained through this process
Weaknesses of qualitative models
Omits small effects or large infrequent effects
Functions often vaguely defined
Loss of detail in space, time, and individual organisms
Presumption of linearity and equilibrium
Time lags not explicit
QUALITATIVE MODELS
PRECISION
GENERALITY REALISM
PRECISION
Richard Levins
1966
GENERALITY REALISM
STATISTICAL MODELS
PRECISION
GENERALITY
REALISM
MECHANISTIC MODELS
Approaches to Complexity
“Making the simple complicated is
commonplace; making the complicated
simple, awesomely simple, that’s creativity.”
(Charles Mingus)
Thanks to:
Jeff Dambacher (CSIRO Mathematics, Informatics, and Statistics),
Sarah Metcalf (Murdoch University),
Pascal Perez (University of Wollongong)