Voting Behaviour and Power in Online Democracy: A Study of LiquidFeedback in Germany's Pirate Party

TL;DR

This study analyzes the emergence and influence of super-voters in Germany's Pirate Party's online liquid democracy platform, revealing that super-voters have high theoretical power but tend to stabilize rather than disrupt the democratic process.

Contribution

The paper introduces a new class of power indices that incorporate observed voting biases, improving prediction accuracy over existing measures.

Findings

Super-voters possess high theoretical power but use it judiciously.

Super-voters often support proposals aligned with the majority, stabilizing the system.

New power indices outperform state-of-the-art measures in predicting voting outcomes.

Abstract

In recent years, political parties have adopted Online Delegative Democracy platforms such as LiquidFeedback to organise themselves and their political agendas via a grassroots approach. A common objection against the use of these platforms is the delegation system, where a user can delegate his vote to another user, giving rise to so-called super-voters, i.e. powerful users who receive many delegations. It has been asserted in the past that the presence of these super-voters undermines the democratic process, and therefore delegative democracy should be avoided. In this paper, we look at the emergence of super-voters in the largest delegative online democracy platform worldwide, operated by Germany's Pirate Party. We investigate the distribution of power within the party systematically, study whether super-voters exist, and explore the influence they have on the outcome of votings conducted online. While we find that the theoretical power of super-voters is indeed high, we also observe that they use their power wisely. Super-voters do not fully act on their power to change the outcome of votes, but they vote in favour of proposals with the majority of voters in many cases thereby exhibiting a stabilising effect on the system. We use these findings to present a novel class of power indices that considers observed voting biases and gives significantly better predictions than state-of-the-art measures.