Communication, Persuasion, Disagreement

Bob Huckfeldt & Paul Johnson
Indiana University
Feb. 12, 2000

Axelrod Culture Model

Grid of "villages" (Agents)
Villages have Features (t1,t2,t3...tN)
Truncated VN-neighborhoods
Step:

Randomly select an agent
Randomly select a neighbor
interact with prob= similarity
copy a randomly chosen feature

Swarm Implementation

http://www.swarm.org
Current development version 2.1.38 (2001-02-07)
This model written in Objective-C
Technical highlights

multi-agent cells
advanced scheduling concepts

Autodoc output:
http://lark.cc.ku.edu/~pauljohn/Swarm/MySwarmCode/Opinion/Documentation

Multi-Agent Grid

This allows incorporation of:

parochialism
agent movement

Measurement Complications

Axelrod used "regions" and "zones"

New Measures: Aggregate

Entropy: objective indicator of diversity

global: diversity of opinion profiles
feature: diversity of opinion on a feature

Opinion distributions
Levels of change/interaction

New Measures: Individual-level

Agents keep personal records on their perceptions

Acquaintance: proportion of others with whom they have at least one feature in common.
Harmony: proportion of features shared with cquaintances.
Identicality: proportion of acquaintances with whom one shares all features.

Display complications

Grid cells now may have multiple members.

Solution: colors indicate summary indicators of opinions of agents within a cell.

Axelrod replication 1: typical outome
Axelrod replication 2: diversity?

Parochialism doesn't do much

Parochialism: probability of choosing discussant from own cell.

Parochialism = 0.50

Features = 5

    Features = 10 stage 1
                     stage2
                     stage3
                     stage4

Parochialism=0.99 same result

Multi-Neighborhood Model

Agents from separate Nhoods might interact within the workplace.

Geographical dispersal allows

exposure
isolation

Importance of "Scheduling" concepts.

Agent movement allows them to:

find new contacts
be found by others
or not!

Our strategy:

divide "day" between homegrid and workgrid
agents move, look for discussants at time of their choosing

Implementation

Dynamic Decentralized Scheduling (various)

Baseline model: Scheduling within Axelrod paradigm

Same result with 5 neighborhoods

snap1 snap2 snap3 snap4 snap5

Variations on Rules

Interactions rules

Coleman model: interact even if there is nothing in common

Select discussant from available agents (various)

Adjustment rules

Survey the "friends": adopt a new rule if 1/2 of acquaintances with whom you share 1/2 of features also hold new view

favorite figure

Compare 5 nhood model
snap1 snap2 snap3 snap4 snap5

Can both agents adjust opinion? (various implementations)