Three methodological issues for system dynamics practice

System Dynamics – three

methodological considerations

Rationality, theory/observation link, “3Ps” in modelling

Andreas Größler

Radboud University Nijmegen, the Netherlands

Issues discussed in this lecture

1. Rationality: perfect vs. bounded rationality, rationality in the model and

the modelling process

2. Theory/observation link: inductive vs. deductive ways to do research,

„Wheel of science“

3. 3Ps in modelling: policy, politics, and polity

Rationality

Rationality =

1. Reasonable, based on reason

2. In an economic sense: choice amongst decision alternatives which

maximises the utility of the decision-maker (with respect to his/her

preferences)

Mindless

behaviour

Perfect

rationality

Optimal decisions are improbable

Real decision situations are characterised by complexity and uncertainty

• In general, optimal decisions are not possible

• absolute (or perfect) rationality changes to bounded, intended rationality

How is rationality measured?

Absolute rationality

– Result counts; it is optimal

– Decision process is

determined by the optimal

outcome

Bounded rationality

– Result is optimal only by chance;

in all other cases it is at best

satisfying regarding an aspiration

level

– Important: decision process and

decision rules

– “procedural rationality”

Bounded rationality and SD

Servo-mechanism theory,

but not: system dynamics

Advances in decision making,

but not: bounded rationality

Literature review

• Morecroft (1983): bounded rationality implicitly embedded in SD models • Morecroft (1985): bounded rationality should be represented in decision

models • Sterman (1987): expectation formation is boundedly rational • Sterman (1989): misperception of feedback as one component of bounded

rationality • Radzicki (1990): institutional economics should use SD to model bounded

rationality • Lane (1994): relevant modelling must include boundedly rational decision-

making • Kampman & Sterman (1998): effects of market mechanisms on outcomes

from boundedly rational behaviour • Dyner & Franco (2000; 2004): modelling bounded rationality in the energy

world

Three starting-points to examine rationality in system

dynamics

Rationality ...

… when creating a model

… in the model‘s structure

… when using the model

process content

Ideal model development process

afte

r F

orr

este

r 1

99

4

Describe

the

system

Convert

description

to level

and rate

equations

Simulate

the model

Design

alternative

policies and

structures

Educate

and debate

Implement

changes in

policies

and

structures

step 1 step 2 step 3 step 4 step 5 step 6

Rationality in model development

Model development:

• Frequently does not follow a formal process

• Depends on skills of modeller

• Objective: Modelling of real, not optimal systems:

• Bounded rational description of bounded rationality (br2)

Importance of validation!

Vicious circle model/reality

Bounded rationality

in the problem area

Difficulty of

knowledge elicitation

Complexity of

modelling process Bounded rationality within

the modelling process

Quality/utility

of model

–

+

+

+

– The utility of “good”

models and the difficulty

of modelling them

Complexity of

problem domain +

–

Bounded rationality in the model structure

• Boundedly rational behaviour of real actors must be replicated in the

model structure/policies (“premise description”)

• In particular, in information flows: wrong relations (functions) or

missing links between variables

• Material flows determined by physical characteristics

• No explanation of reasons for bounded rationality

goods on stock

goods delivered to

customer shipment

customer order rate

fulfilment ratio (forgetting)

shipment delay

remoteness factor

… …

How to handle bounded rationality

• Habit, routines, and rules of thumb

• Managing attention

• Goal formation and satisficing

• Problem decomposition and decentralized decision making

afte

r S

term

an

20

00

Need to be represented in model

Filter in policies

Level

rate 1. Cognitive limitations

2. Operating goals, rewards

and incentives

3. Information,

measurement and

communication systems

4. Organisational and

geographical structure

5. Tradition, culture, folklore

and leadership

1 2

3

4

5

afte

r M

ore

cro

ft 1

99

4

Critical distinction

Which issues are necessary

abstractions in the model

development process? Which issues are simplified

in accordance with

artefacts of bounded

rationality occurring in

reality?

Which issues are

mistakenly simplified by

modeller?

Rationality when using a model

• Using a model means simulation experiments = scientific approach

• Goal: Improved, more robust policies, i.e. less boundedly rational

decisions because simulation should overcome cognitive

limitations of humans

Improvement of policies

• Frequently, no structural changes, only variation of parameters =

acceptance of bounded rationality

• Design of robust policies requires changes in model structure and,

hence, in organisational structure

• Bounded rationality:

• Negative influence in the modelling process and on simulation

experiments

• But: Model structure should represent bounded rationality of real

decisions

Summary (so far)

Formal Model Real World Problem

modelling

simulation

incorporation of

bounded rationality

learning to mitigate

bounded rationality

bounded rationality

of model developer

bounded rationality

of model user

Bounded rationality in model structure: An example

Inventory model, Lyneis 1981

Supplier

Production

Personnel

Customers demand

parts

ordered

parts

received

desired

production

capacity

personnel

shipment

Supplier

Make-to-order

Auftragsbestand

von Zulieferer

bestellrate der teile

lieferzeit teile

zulieferer

rüstverzögerung

+

durchschnitt produktion

des zulieferers startet

-

gewünschte

produktionsrate des

zulieferers

auftragsbestand zu

erledigen

+

ZEIT UM

AUFTRAGSBESTAND

ZU ERLEDIGEN

-

durchschnittliche

bestellrate

+

gewünschte

produktionskapazität

des zulieferers

+

+

gewünschter

auftragsbestand des

zulieferers

KAPAZITÄT

DES

ZULIEFERERS

anpassung der

produktionskapazität

des zulieferers

ZEIT UM

PRODUKTIONSKAPAZITÄT

ANZUPASSEN -

+

+

-

kapazitätsauslastung

des zulieferers

+

-

TABELLE FÜR

KAPAZITÄTSAUSLASTUNG

DES ZULIEFERERS

MIMIMUM

ZULIEFERER

RÜSTVERZÖGERUNG

+

-

+

+

produktionsverzögerung

des zulieferers +

erhaltene lieferzeit

teile

ZEIT UM LIEFERZEIT

TEILE ZU ERHALTEN

+

Bestellte

Teile produktion des

zulieferers startet

+ +

eingangsrate

+

GLÄTTUNGSZEIT

BESTELLRATE DES

ZULIEFERERS

Supplier as ‘homo oeconomicus’

• No capacity restrictions

• Infinitely fast reactions

• Complete knowledge about future (certainty)

• Result: In each period produces exactly the amount of goods that

is demanded optimal solution

Supplier in the Lyneis model

• Only one information cue used to decide about capacity and production: order rate of producer

• Order rate is smoothed to filter out peaks • This figure serves as a prognosis value for future order rates • Very inexact estimation bounded rationality

More robust policies for the supplier

• Shorten reaction times if possible and useful

Change of parameters

• Use other processing rules, e.g. investment algorithms instead of

permanent capacity adjustment

Change of functional relations

• Consider more information, e.g. expected order rate at producer, data

about business cycles, seasonal effects

Change of structural linkages

Bounded rational policies can be dangerous

afte

r S

term

an

20

00

Demand

Capacity

Utilisation

Capacity

Price

Market Share

Competitor

Price

Industry Demand

+

+

+

- +

-

+

Fill the Line

Competitor

Market Share

Competitor

Demand

Competitor

Capacity Utilisation

Competitor

Capacity

+

+

+

+

-

-

+

Fill the Line

Price War

“The road to hell is paved with good intentions”

Local rationality leads to crisis, catastrophe, bankruptcy

But: can success emerge from boundedly rational behaviour?

Locally bounded, but globally successful

Agent

10

.

.

.

Agent

1N

Agent

20

.

.

.

Agent

2M

Agent

11

Agent

21

Agent

30

.

.

.

Agent

3L

Agent

40

Agent

4K

Agent

31

Agent

41

Level 1 rate 1 rate 2

aux 1 aux 2

CONST 1

Level 2 rate 3 rate 4

CONST 2

aux 3

Theory/observation link

30

Goals of Human Inquiry

• Making sense of the world

• Common sources: tradition/authority and personal experience

(Asch’s experiment) prove that the earth revolves around the

sun!

• Explain and predict: why? and what?

• Prediction without explanation is possible; explanation often leads

to prediction

• Predict and control ( interventions)

Shortcomings of Human Inquiry

• Inaccurate observations (visual puzzles)

• Overgeneralizations (“all...are...”)

• Selective perception/observation (looking for confirmation)

• Illogical reasoning (“the exception that proves the rule”, gambler’s

fallacy)

Science

• Making sense of the world in a specific way • Knowledge in terms of statements about reality • Generation of new knowledge through systematic (scientific)

research • Objective: describe and explain ‘reality’ (knowledge) pre-

/modern/post-modern • Our knowledge materializes in statements about that reality (laws =

observed regularities, not individual exceptions) • Research uses methodology/methods • Management sciences: research and intervention

The Foundations of Social Science

• The Charge of Triviality (Darwin: “fool’s experiment”)

• Social Regularities Aggregates, Not Individuals

• What About Exceptions? (probabilistic predictions) The collective

actions and situations of many individuals.

• People Could Interfere (if “irregular” behavior becomes

commonplace, new theories are needed) Focus of social science

is to explain why aggregated patterns of behavior are regular even

when individuals change over time or how the regularities change.

Social Sciences: Issues

• Finding universal laws is problematic

• A couple of reasons:

• Complexity

• Researcher effect

• Research changes reality

• We often end up with statements like “in general”, “in principle”,

“with a high likelihood”, “under this and that condition”, …

• discussion in the beer game

The Links Between Theory and Research

• Deductive Model – research is used to test theories.

• Inductive Model – theories are developed from analysis of data.

• The Traditional Image of Science – The deductive model of scientific

inquiry begins with a sometimes

vague or general question, which is

subjected to a process of

specification, resulting in hypotheses

that can be tested through empirical

observations.

Social versus natural sciences

• Differences in research object – reflexivity/reactivity

• Social sciences similar objectives? (interventions?)

• Idea of unity of sciences (logical positivism)

The wheel of science

Theories

Hypotheses

Observations

Empirical

generalizations

Ind

ucti

on

Ded

ucti

on

Types of theories, types of models

Range of

theory

Goal of

theory

content structure

Explaining…

grand

theory

midrange

theory

minor

theory

“3Ps” in modelling

What system dynamics wants to achieve…

• Policy design/ Policy making: decision processes that convert

information into action (Forrester, 1994). <-> decisions

• Policy design is an analytical/cognitive/rational task

• However, be aware of a too mechanistic/rationalistic view of

organisations

• Therefore, consider politics and polity

• Politics: games played on the self-interest of people

• Polity: institutional structures in which politics/policies take place

The 3Ps and system dynamics models

Problem

articulation

Dynamic

hypotheses

Model

formulation

e.g. Policies

Model

testing

Policy*

formulation&

evaluation

Politics

Polity

Politics*

Polity*

The 3Ps and the system dynamics modelling process

Problem

articulation

Dynamic

hypotheses

Model

formulation

e.g. Policies

Model

testing

Policy

formulation&

evaluation

Politics Polity

Policies

The 3P‘s The modelling process determines

scope

determines

implementation

changes due to

model results

and modelling

process

References

Andreas Größler, Peter Milling and Graham Winch (2004): Perspectives on

Rationality in System Dynamics – A Workshop Report and Open Research

Questions, System Dynamics Review, 20(1), pp. 75–87.

Andreas Größler (2008): System Dynamics Modelling as an Inductive and

Deductive Endeavour, Systems Research & Behavioral Science, 25(4),

pp. 467–470.

Andreas Größler (2010): Policies, Politics, and Polity, Systems Research &

Behavioral Science, 27(4), pp. 385–389.

45