makeAgents: Agents are constructed with a bias on their binary decision...

In oxacclab/adviseR: Advice Exchange Network Modelling

Description

Agents are constructed with a bias on their binary decision (constant over time) and a set of weights governing how seriously the advice of other agents is taken (modified over time).

Usage

makeAgents(
  n_agents = n_agents,
  n_decisions = n_decisions,
  bias_mean = 0,
  bias_sd = 1,
  sensitivity_mean = 1,
  sensitivity_sd = 1,
  trust_volatility_mean = 0.05,
  trust_volatility_sd = 0.01,
  bias_volatility_mean = 0.05,
  bias_volatility_sd = 0.01,
  confidence_slope_mean = 1,
  confidence_slope_sd = 0,
  weighted_sampling_mean = 0,
  weighted_sampling_sd = 0,
  starting_graph = NULL
)

Arguments

n_agents	number of agents to create
n_decisions	number of decisions
bias_mean	the mean for the agents' bias distribution (agents' biases are drawn from normal distributions with mean +/- bias_mean). Fed into a sigmoid function and the capped to between 0 and 1. Represents the prior probability that the answer is 1.
bias_sd	standard deviation for the bias distribution
sensitivity_mean	mean for agents' sensitivity
sensitivity_sd	standard deviation for distribution of agents' sensitivity (mean is 1)
trust_volatility_mean	the mean volatility of agents' trust
trust_volatility_sd	standard deviation
bias_volatility_mean	the mean volatility of agents' biases (move this proportion towards the final decision value from current bias at each step)
bias_volatility_sd	standard deviation
confidence_slope_mean	the mean of the distribution from which agents take their slopes for the sigmoid function mapping continuous evidence to a probability of a categorical decision.
confidence_slope_sd	standard deviation
weighted_sampling_mean	a non-zero value means agents choose who to seek advice from according to how likely they are to trust the advice. The weights are raised to the power of this value (so values > 1 make source selection more pronounced than advice weighting, and values < 1 make source selection less pronounced than advice weighting). Negative values will make agents actively seek out those they do not trust for advice.
weighted_sampling_sd	standard deviation
starting_graph	single number, vector, or n_agents-by-n_agents matrix of starting trust weights between agents. Coerced to numeric. Can also be a function taking the first generation of the agents tbl as an input and returning an n-by-n matrix of trust values between 0 and 1, where n is the number of agents, 0 represents completely untrustworthy, .5 random, and 1 completely trustworthy.

Details

the agents tibble is an n_agents*n_decisions by 12 table with

• "id"The agent's identifier

• "decision"The decision number

• "sensitivity"The agent's ability to do the task

• "trust_volatility"How quickly the agent's trust updates

• "bias_volatility"How quickly the agent's bias updates

• "weighted_sampling"How heavily trust governs advice sampling behaviour

• "bias"The agent's (initial) bias

• "truth"The true state of the world (same for a given decision for all agents)

• "initial"The agent's initial estimate of the truth

• "advisor"ID of the agent whose advice is received

• "weight"The weight assigned to the advice by the agent

• "final"The final decision of the agent

Value

list( agents = tibble of agents' decisions, advice, etc. at each time point graphs = list of agents' trust matrix for each time point )

oxacclab/adviseR documentation built on Oct. 7, 2021, 8:05 p.m.