Zum Inhalt

Participation Dilemma and Group-Relative Learning Concept

The thesis baseline uses a group-relative participation signal because the model's incentive structure is not purely individual.

1. Participation Dilemma Built Into the Model

The model intentionally combines:

  • participation fee paid only by participants
  • collective reward/punishment mode based on the Puzzle Quality Gate
  • reward magnitude tied to preference alignment
  • both fees and rewards are relative to agents assets

Participation cost is individually concentrated, quality sign is socially shared, and within preference groups the reward component is largely shared while the fee remains individual. That creates a structured free-rider tension.

2. Why Fees and Rewards Are Not Absolute but Relative to Assets

This choice reflects the relative nature of effort and reward in the model. The model does not attempt to represent direct real-world wealth-power effects. Instead, it models motivational resources and reward pressure in relative terms. Absolute differences still remain analytically visible and comparable.

3. Why Pure Individual Reward Learning Is Weak Here

If participation learning uses only immediate personal reward-like signals, the effective differentiating information can collapse toward fee effects, reducing meaningful group-competition structure.

For this thesis trajectory, learning is therefore framed around relative group performance under participation cost.

4. Group-Relative Party Mode (Baseline)

The thesis baseline uses participation_signal_mode=group_relative_delta_rel_party.

Definitions for each step:

  • mu_g: mean election_delta_rel among eligible agents in group g
  • mu_bar: mean of mu_g across groups with eligible members
  • r_g = w_g * (mu_g - mu_bar)
  • w_g = n_g / (n_g + participation_signal_group_shrink_k)

Per-agent signal:

  • signal_i = clip(r_g + fee_component_i, +/- participation_signal_clip)
  • participant fee term: fee_component_i = - participation_signal_fee_weight * fee_rel_i
  • abstainer fee term: fee_component_i = 0

5. Update Direction for Participants and Abstainers

In this baseline mode, for every eligible agent i (participant or abstainer):

  • q_i <- q_i + participation_alpha * signal_i

So both participants and abstainers receive the same group-relative directional component r_g. Only participants receive the additional negative fee component.

So the group-level direction can move both subpopulations similarly, while participant updates are damped or locally reversed when fee salience dominates.

6. Why This Fits the Thesis Question

The thesis asks how voting rules affect participation dynamics. This learning design increases sensitivity of participation updates to collective comparative outcomes, making rule effects on participation more observable and interpretable.

7. Scope Boundary

This is a stylized party-relative learning rule, not a claim about how real voters literally update behavior.