- Download page: https://ees4760.jgilligan.org/downloads/hoopoe_class_21/
- Zip file https://ees4760.jgilligan.org/models/class_21/class_21_models.zip containing Wood Hoopoe breeding model:
- NetLogo model: https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_class_21.nlogo
- ODD: https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_odd.pdf
- NetLogo model with alternative strategies: https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_strategies.nlogo

- How to use models to run experiments?
- Strong inference (John Platt)
- Identify traits (individual behaviors) that give rise to multiple macroscopic patterns

- Identify alternative traits (hypotheses)
- Implement traits in ABM
- Test and compare alternatives:
- How well did model reproduce observed patterns?
- Falsify traits that did not reproduce patterns

- Repeat cycle as needed. Revise behavior traits, look for additional patterns, etc.

- Traders establish buying and selling prices
- If someone offers a price \(\ge\) selling price, trader sells.
- If someone offers to sell for \(\le\) buying price, trader buys

- Match traders:
- If traders \(i\) and \(j\) have \(P_{i,\text{sell}} \le P_{j,\text{buy}}\), then transaction occurs.

- Agent sets random buying and selling price
- If \(P_{i,\text{buy}} > P_{i,\text{sell}}\), then trader \(i\) will lose money.

- Random buying and selling price with constraint: \(P_{i,\text{buy}} < P_{i,\text{sell}}\).

- Minimal-intelligence agent was better than zero-intelligence
- Zero-intelligence produced wild price fluctuations
- Minimal-intelligence reproduced observed pattern of rapid price convergence
- Minimal-intelligence also reproduced observed effects of price-ceiling.

- But simple models had limits:
- Observed volatility of lower-end prices was not reproduced by models
- As experimental markets got more complicated, human traders did worse, but models did
worse.*much*

Using zero-intelligence as a baseline, the researcher can ask: what is the minimal additional structure or restrictions on agent behavior that are necessary to achieve a certain goal.

- Experimental subjects move avatars on screen to harvest tokens

(like simple video game) - Players compete to get most tokens
- Tokens grow back at some rate
- Patterns:
- Number of tokens on screen over time
- Inequality between players
- # tokens collected in first four minutes
- Number of straight-line moves

- Näive model: (random) Moves randomly
- Näive model: (greedy) Always goes to nearest token
- Clever model:
- Prefers nearby tokens
- Prefers clusters of tokens
- Prefers tokens straight ahead
- Avoids tokens close to other players

- Näive models do not match any of the four patterns.
Ran clever model 100 times for each of 65,536 different combinations of parameters that characterize preferences.

- Only 37 combinations of parameters matched all four patterns in data.
- Patterns 2 and 3 are seen for most parameter values
- Patterns 1 and 4 seen less frequently
- Therefore:
- Patterns 2 and 3 are built into the structure of the game.
- Patterns 1 and 4 may give insight into human behavior.

https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_class_21.nlogo https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_odd.pdf

- Groups occupy spatial territories
- One
**alpha**of each sex in a territory - Only alpha couple reproduces
- If alpha dies, oldest subordinate of that sex becomes alpha
**Scouting forays**- Subordinate adult leaves territory
- If it finds territory without alpha, it stays, becomes alpha
- Otherwise, returns home
- Risk of predation (death) is high on scouting forays

- Alpha couple breeds once a year, in December

Characteristic group size distribution (adults)

- Average age of birds on scouting forays is younger than

average age of all subordinates. Scouting forays most common April–October

- Start simple:
- One-dimensional world
- One tick = one month
- Every tick, bird has 1% chance of dying (0.99 probability to survive)
- Scouting forays have 20% chance of death (0.80 probability to survive)
- Adult subordinates go scouting at random (50% probability each tick)

- Does model reproduce patterns?

https://ees4760.jgilligan.org/models/class_21/wood_hoopoe_strategies.nlogo