Running Simulations

MACS 30000: Perspectives on Computational Analysis
University of Chicago

Annoucement

  • Assignment 2 due on this Sunday
  • Grades for Assignment 1 will be released no later than tomorrow morning

The story begins with

Schelling, Thomas C. “Dynamic models of segregation.” Journal of mathematical sociology 1.2 (1971): 143-186.

Why simulations?

  • Impossibility of running some social experiments
  • Studying non-equilibrium states of social processes
    • Relaxing the rational-actor hypothesis
    • Modeling heterogeneous agents
    • Understanding time dynamics

Studying emergence

  • Some macro-level social phenomena are emergent.
  • Examples:
    • Wealth inequality
    • Residential segregation
    • Social revolution

Complexity (from Page 2015)

  • Social systems are oftentimes complex (requires a lot of information to describe).
  • They are neither completely ordered nor random: BOAR (between ordered and random)
  • The exact sequence of events are difficult to reproduce. (We are only living in one of all possible states.) DEEEP (difficult to explain, evolve, engineer, or predict); subject to initial conditions and path dependence.
  • Individual behaviors are interactive and interdependent.
  • Individuals learn and adapt.

Example

Complexity oftentimes occurs in-between.

To explain it is to grow it.

Agent-based model (ABM)

  • Agents with heterogeneous internals states interact with an environment (and other agents) with behavioral rules
  • They learn, adapt, reproduce, and perish.
  • Social scientists who run the simulations study the macro-level outcomes such as:
    • population growth
    • price dynamics
    • wealth distribution
    • social networks
    • emergence of institutions

The Sugarscape Model: Epstein & Axtell (1996)’s artificial socieities

In the simplest model, each agent

  • has a vision
  • moves
  • harvests sugar
  • consumes sugar
  • dies if unable to collect enough sugar And the environment grows back to its capacity according to some rules

Results

Emergence of wealth inequality

Emergence of networks

Migration

Seasoned migration

Migration with pollution

You can also simulate artifical society with real data

Anasazi study (Dean at al 2006)

  • Anasazi: a pre-historic society in Long House Valley, northeastern Arizona, 1800 B.C. - A.D. 1300
  • Extensive archaeological data, which has been digitally archived
  • Topographically bounded and self-contained

Simulation rules

  • Unit of obsevation: household of five
  • Independent household agent characteristics: age, location, and grain stocks
  • Shared characteristics: death age and nutritional need
  • Environment: maize growth capacity estimated from environmental data
  • Every year, household agents harvest the amount of grain determined by the capacity at their locations plus some random factors

Simulation rules

  • They consume their nutritional requirements, store the extra if there is a surplus, and die or emigrate if there is a deficiency.
  • They also move their locations according to their estimates of the land capacity.
  • Households can also fission with some defined probability.

Results

  • With simple simulation rules, the researchers are seemingly able to regenerate the history.
  • They also find that environmental factors alone cannot explain the final extinction and argue that they can still learn from the models even if the models cannot fully approximate reality.

Results

Further studies and critiques

  • The parameter space is two huge for exhausting all combinations. The results of simulations could differ greately due to different parameter specifications (Stonedahl & Wilensky 2010).

Further studies and critiques

  • The success of the ABMs can be mostly explained by adjusting some parameters on carrying capacity. The endogenous processes do not seem to matter a lot excepting for smoothing the curves (Janssen 2009).

Cool stuff in 2023

Activity for Thursday

  • Compare and contrast DellaPosta et al. (2015) and Goldberg & Stein (2018).
    • What are the common theoretial/empirical puzzles that these two studies both try to address?
    • Why can’t the problem be easily answered by am empirical study?
    • How are the perspectives of the two papers different?
    • How do the authors respectively model their agents to validate their theoretical hunch?
    • In your opinion, which study better approximates reality?
    • After reading these two papers, have you learned anything about cultural diffusion?

Group mental exercise

  • Propose one social science research question that can hardly be answered by empirical data alone but can be potentially answered by some simulation studies.
    • Explain why the problem is empirically hard to study in the first place and why a simulation study could help generate unique insight.
  • Try to briefly sketch a research design if you can.
    • How would you model the agents?
    • Given the complexity of the problem, why do you think that your simulations can reasonably approximate reality?
    • Or, do you believe that even if they can’t, you can still learn a good deal from them?

Discussion groups

0 Huanrui / Emma / Tianle
1 Jiazheng / Yuhan / Max
2 Lorena / Kuang / Zhuojun
3 Daniela / Anny / Yue
4 Pritam / Cosmo / Thomas
5 Andy / Agnes / Kexin
6 Abbey / Ertong / Tian