A Model Is A Model Is A Model

EES 4760/5760

Agent-Based & Individual-Based Computational Modeling

Jonathan Gilligan

Class #2: Thurs. January 11 2018

A Model Is A Model Is A Model

Modeling

  • There are many “modelings”:
    • Agent-Based Models
    • Mathematical Models
    • Statistical Models
    • System-Dynamics Models
    • Discrete-Event Models
    • Stochastic Dynamic Models
    • Physical Models

Physical Model of the Economy

MONIAC: A physical model of the British national economy

(Photo: Wm. Vandivere, Fortune, March 1952, p. 100)

Agent-Based Modeling (ABM)

  • Two elements:
    1. Agent-based
    2. Modeling
  • Certain principles apply to all kinds of modeling
  • First, consider modeling
  • Then consider what distinguishes agent-based modeling from other kinds.

What Is A Model?

  • Definition (first try):
    • A model is a simplified representation of reality
  • Why do we simplify?

Modeling

  • Developing a model:
    • Problem solving under constraints
  • Most important constraints:
    • Incomplete information
    • Lack of time
    • Lack of resources (people, money, computing power, etc.)

Example

  • You bought a six-pack of a tasty beverage last night,
    • but when you get home this evening, you realize that you forgot to put it in the refrigerator.
  • So you put it in the fridge, and now you want to know, without trial and error, when it will be cool enough to drink.
  • How do you approach the problem?

Heuristics

Heuristics

  • Mental shortcuts
  • Rules of thumb that experience has shown to be useful.
  • When solving problems under constraints, apply heuristics in modeling:
    • Simplified representations

Typical Heuristics

  • Rephrase the problem
  • Draw a simple diagram of the system
  • Imagine that you are inside the system
  • Identify essential variables
  • Identify simplifying assumptions
  • Use “salami tactics”: slice space and time

What Is A Model

  • Definition (second try):
    • A model is a purposeful (simplified) representation
  • Modeling is something we all do all the time because we never have enough data and time!

  • Thinking = problem-solving = modeling

What Is A Model?

  • Modeling adaptive behavior means trying to model the models used by adaptive agents (plants, animals, humans, organizations, etc.)
  • A model is a model is a model

Is Modeling Essential?

Is Modeling Essential?

  • When trying to solve a problem, we keep asking ourselves, “is this aspect of the real system essential for solving my problem?”
  • How can we know whether something is essential?
    • We cannot know
    • In science, we keep developing the model to test our assumptions

Example: Model A Forest

  • Without a clearly stated question or problem we cannot formulate a simplified representation.
    • We don’t know the purpose of the model
  • The strategy:

    Model first, then think about what problems we can solve with the model

    does not work!
  • Forest model:
    • Timber extraction
    • Ecosystem preservation
    • Forest fires

Example: Checkout Queue

  • Your purpose:
    • Minimize waiting time
  • Manager’s purpose:
    • Minimize waiting time of all customers
  • Manager’s solution:
    • Single queue for all customers
      • Airports, banks, etc.

Lessons for Agent-Based Modeling

Lessons for Agent-Based Modeling

  • ABM requires some specific techniques (programming, math, statistics)
  • But general modeling principles apply.
    • Scientific modeling explicitly states heuristics, simplifying assumptions
    • Use math & computer logic to rigorously explore consequences of assumptions

Lessons for Agent-Based Modeling

  • We must start with a clearly formulated research question
  • We need to simplify
  • Iterative process:
    • Formulate question
    • Create simplified representation
    • Implement model as program
    • Test program
    • Analyze output
    • Start over with modified question/model/program/etc.
  • Modeling cycle

The Modeling Cycle

The Modeling Cycle

Modeling Cycle

Modeling Cycle Tasks

Formulate the Question

  • Question or problem serves as filter for what to include in the model.
  • Modeling the system first and then specifying the question does not work

Assemble Hypotheses

  • We need a conceptual (often verbal, graphical) model of how the system works and what the answer is.
  • This conceptual model can be based on: empiricial experience, theory, feeling
  • Discuss and revise the conceptual model thoroughly, but not forever.
    • It can’t be tested in your head!

Modeling Cycle Tasks

Choose Model Structure

  • What are the model’s entities?
    • How are they characterized (state variables)?
  • How do you represent the environment?
  • What are temporal and spatial resolutions and extents?

Implement the Model

  • Write down equations and/or implement model as computer program
  • Choose appropriate software platform/system

Modeling Cycle Tasks

Analyze the Model

  • Perform controlled experiments to understand your model
  • Design & analyze simulation experiments just like real experiments
  • This is the hard part (95% of the time)

Communicate the Model

  • Like lab protocol: Model development has to be documented
  • Keep a notebook of what y ou do.
  • Keep old versions of your model
    • Name files model_1.nlogo, model_2.nlogo, etc.
    • Or use revision-control software (git, mercurial, etc.)
      (See “Reading Resources and Computing Tools” handout)
  • Final documentation should enable peers to fully understand and re-implement model (ODD specification) (More on this next week)

The Modeling Cycle

Modeling Cycle

Example Models

Examples of Agent-Based Model Research at Vanderbilt

  • Impact of land-use change on Brazilian ecosystems
  • Spread of solar-roof systems in California
  • Adaptation to drought by rice farmers in Sri Lanka
  • Interaction of land-use and sea-level rise in Bangladesh
  • Interaction of environmental change and population migration in Bangladesh
  • Co-evolution of mega-herbivores and steppe ecosystems at Pleistocene-Holocene boundary
  • Predicting traffic congestion for navigation apps
  • Impact of Nashville gentrification on mobility & access to mass transit
  • Can prediction market affect belief in climate change?
  • Teaching K-12 science

Other Examples of Agent-Based Model Applications

  • Predator-prey interactions
  • Preserving viability of threatened species
  • Interaction of public belief in global warming, engineering projects, and future vulnerability of coastal communities.
  • Impact of natural disasters on cities
  • Designing effective political institutions
  • Designing evacuation routes from buildings
  • Predicting and managing disease epidemics
  • Mechanisms of septic shock (bacteria in human body)
  • Developing strategies for responding to mass shootings
  • Effect of opium trafficking on Taliban insurgency in Afghanistan