Simulation Models as a Research Method

1
Simulation Models as a Research Method

• Professor Alexander Settles

2
Research Methodology - Simulation

• Simulation as a research tool
• Research in simulation
• Focus here is on simulation of discrete event
  dynamic systems

3
Social Simulation

• Most social science research uses some kind of
  theory or model
• Theories are generally stated in textual form
• But some are represented as equations
• Sometimes carry out experiments on artificial
  social systems that would be impossible or
  unethical to perform on human populations
• One advantage must think through your
  assumptions
• Clarity and precision each parameter needs a
  value
• All the detail of the model can be inspected by
  others
• Disadvantage data adequate for estimating all
  parameter values may be hard to get

4
Sociology and Complexity

• The physical world is full of systems that are
  (almost) linear
• But (human) societies have quite unpredictable
  features
• Their characteristics at any one time are
  affected by their past histories (path
  dependence)
• E.g., adoption of 1 of a pair of alternative
  technologies by a society can be greatly
  influenced by minor contingencies about who
  chooses which technology early on
• Human societies, institutions and organizations
  are complex systems
• The behavior of the system as a whole cant be
  understood in terms of the separate behaviors of
  its parts
• Contrasts with reductionist physical sciences

5
Simulation as a Research Tool

• Why simulation?
• An analytical approximation has been developed to
  model some system performance measure.
• The development of the approximation requires
  simplifying assumptions/approximations.
• The conjecture is that the analytical model is
  still a reasonable representation of the real
  system.
• Simulation is being used to support or refute
  this conjecture.

6
Simulation as a Research Tool

• Are the assumptions applied in the simulation
  clearly stated?
• Distributions used.
• Operational protocols, e.g., blocking, etc.
• Correlation?
• Can you simulate the same system?
• Steady State vs. Terminating
• Number of runs
• Length of runs
• Some models take a long time to settle down

7
Simulation as a Research Tool

• Verification validation
• Mainly applies to studying a real system or a
  detailed representation
• How was this conducted?
• Results compared to an existing system?
• Comparisons made to existing analytical results?
• Extreme cases tested?

8
Simulation as a Research Tool

• Experimental design
• Experimental design?
• Random systems?
• The importance of this depends on the way the
  simulation was used
• If simulating to understand a system and gain
  insight, these issues become more important

9
Methods of simulation

• System dynamics
• Behavior of a system with complex causality and
  timing
• System of intersecting, circular causal loops
• Stocks that accumulate and dissipate over time
• Flows that specify rates within system
• Inputs to a system of interconnected causal
  loops, stocks, and flows produce system outcomes

10
System Dynamics Research Tools

• Add causal loops
• Change mean of flow rates
• Change variance of flow rates

11
System Dynamics Research Questions

• How do organizations undergo fundamental change?
• When do small interruptions create major
  catastrophes?
• What conditions create system instability?

12
NK fitness landscapes

• Speed and effectiveness of adaptation of modular
  systems with tight versus loose coupling to an
  optimal point
• System of N nodes, K coupling between nodes
• Fitness landscape that maps performance of all
  combinations

13
NK Fitness landscape

• (S, V, f)
• S set of admissible solutions,
• V S ? 2S function, neighborhood
• S ? IR fitness function.

14
Key Assumptions

• Adaptation via incremental moves and long jumps
• Optimization
• Adaptation of a modular system using search
  strategies (i.e., long jumps, incremental moves)
  to find an optimal point on a fitness landscape

15
NK fitness landscapes

• Vary N and K
• Change adaptation moves
• Add a map of the landscape
• Create an environmental jolt

16
NK fitness landscapes

• How long does it take to find an optimal point
  (e.g., high-performing strategy)?
• What is the performance of the optimal point?
• What is the optimal strategic complexity?
• How does cognition improve experiential learning?

17
Genetic algorithms

• Adaptation of a population of agents (e.g.,
  organizations) via simple learning to an optimal
  agent form

18
Genetic algorithms

• Adaptation of a population of agents (e.g.,
  organizations) via simple learning to an optimal
  agent form
• Population of agents with genes
• Evolutionary adaptation (v-s-r)
• Variation via mutation (mistakes) and crossover
  (recombination)
• Selection via fitness (performance)
• Retention via copying selected agents

19
Theoretical Logic

• Optimization
• Adaptation of a population of agents using an
  evolutionary process toward an optimal agent form

20
Research Questions

• How does adaptive learning occur within
  bargaining?
• How does organizational learning affect the
  evolution of a population of organizations?
• What affects the rate of adaptation (or learning
  or change)?
• When and/or does an optimal form emerge?

21
Genetic algorithms
22
Cellular automata

• Emergence of macro patterns from micro
  interactions via spatial processes (e.g.,
  competition, diffusion) in a population of agents

23
Cellular automata

• Population of spatially arrayed and
  semi-intelligent agents
• Agents use rules (local and global) for
  interaction, some based on spatial processes
• Neighborhood of agents where local rules apply

24
Research Questions

• How does the pattern emerge and change?
• How fast does a pattern emerge?
• How do competition and legitimation affect
  density dependence?

25
Stochastic processes

• One or more processes by which system operates
• One or more stochastic sources (e.g., process
  elements)
• Probablistic distributions for each stochastic
  source

26
Definition

• A stochastic process is one whose behavior is
  non-deterministic in that a system's subsequent
  state is determined both by the process's
  predictable actions and by a random element.
• Manufacturing process
• Finance asset pricing Markov chain

27
Research Questions

• What is the relationship between exploration and
  exploitation?
• What is the optimal degree of structure?