Liquid democracy in practice: super-voters, low turnout and a theoretical warning

Liquid democracy lets each member of a group either vote directly on an issue or delegate their vote to someone they trust — revocably, often by topic, and in many designs transitively, so that a delegate can pass on the votes they hold. Advocates present it as a way to combine the equality of direct democracy with the expertise of representation. This study reviews what has happened when the idea was used: in the German Pirate Party's LiquidFeedback platform (2010–2013), in a German district's LiquidFriesland experiment (2012–2015), in Google's internal Google Votes, in Democracy Earth's blockchain design, in delegation modules for the Decidim platform, and in the delegated token voting of decentralised autonomous organisations (DAOs). It also reviews the main theoretical results. The practical record is short and consistent. Delegation produces a small number of "super-voters" holding very large voting weights; participation is low and declines after an initial burst; and the tools have survived mainly in organisations that need low-cost internal decisions, not in public government. Theory adds a sobering result: simple, local delegation rules cannot guarantee better decisions than direct voting.

Key findings

• Delegation concentrates power. In the German Pirate Party's LiquidFeedback (499,009 votes on 6,517 initiatives, August 2010 to November 2013), the distribution of voting weight followed a power law; 38 voters held more than 100 delegations. Removing delegations would have changed about one in six outcomes (84.9% unchanged).
• Super-voters used their power sparingly. The same study found that powerful delegates reversed results in only about one in three votes where they could have done, and tended to vote with the majority — but the authors warn that such restraint need not be stable.
• Participation was low. In 2012 about 10,000 of the party's roughly 30,000 members were registered on the system and an estimated 3–5% used it actively. LiquidFriesland registered 583 of more than 80,000 eligible residents in nearly three years; no vote drew more than 19 participants in its last year, and activity stopped in April 2015.
• DAO delegation is even more concentrated. On 1 March 2022, 8 delegates in Compound, 11 in Uniswap and 18 in ENS together held a majority of delegated voting power; a later study of 21 DAOs found that in 17 of them fewer than 10 participants controlled a majority.
• Theory warns against naive delegation. Kahng, Mackenzie and Procaccia (2021) proved that local delegation mechanisms — the kind that capture the spirit of liquid democracy — cannot guarantee to outperform direct voting even when voters delegate only to better-informed voters.
• Normative theory supports modified forms. Blum and Zuber (2016) argue that liquid democracy can beat representative democracy on both epistemic and equality criteria, but only with deliberation, trustee-style delegates and an executive that checks overall consistency.
• Corporate and cooperative uses are the most durable. Google ran an internal liquid-democracy system for about three years; Decidim offers delegation in a module for cooperatives; DAOs use one-step delegation of tokens at scale.

Research question and scope

The question is: what does the evidence show about how liquid (delegative) democracy behaves in practice — who participates, how voting power is distributed, how decisions differ from direct voting, and whether the tools last? The study covers implementations with published data or documentation: party platforms, local government pilots, corporate experiments, open-source civic software and blockchain governance. It reviews theoretical work from political philosophy and computer science where it bears on these questions. It does not cover proxy voting in company law or the long history of representative assemblies, except as background.

Background

The idea is older than the internet. Charles Dodgson (Lewis Carroll) described a scheme in 1884 in which votes could be passed to other candidates. Modern liquid democracy became practical with online platforms. Paulin's (2020) overview dates the first technical systems to 2009: Župa, developed for a student organisation in Slovenia, and LiquidFeedback, built by supporters of the Pirate Party in Berlin. Since then the idea has been developed in two literatures that rarely meet: political theory, asking whether delegation is legitimate and desirable, and computational social choice, asking what delegation does to the accuracy and fairness of collective decisions.

Typical designs share four features: anyone may vote directly; anyone may delegate, globally or by topic; delegation is revocable at any time; and, in most designs, delegation is transitive, so that votes can flow along chains. They differ on whether votes are public (LiquidFeedback made all votes and delegations visible to members), whether a voter's direct vote overrides their delegate, and whether delegation can cross topic boundaries.

Method and sources

This is a document and literature review. For each implementation we sought primary documentation (platform descriptions, code repositories, organisation reports) and independent analysis with data. The LiquidFeedback evidence rests mainly on Kling and colleagues' analysis of complete database dumps; LiquidFriesland on an evaluation summarised by the Konrad Adenauer Foundation, a foundation close to the Christian Democrats, which had criticised the project — its figures are drawn from the district's own evaluation, but its framing should be read with that in mind. DAO evidence comes from blockchain data analysed by researchers at ETH Zurich. Google Votes is documented only in a technical disclosure by its designers, without usage statistics in its abstract. The evidence base is therefore thin, concentrated in a few cases and partly produced by participants or critics.

Finding 1: the Pirate Party and LiquidFeedback

The platform

The German Pirate Party adopted LiquidFeedback in May 2010. Members could create initiatives, which needed a quorum of supporters before going to a vote; competing initiatives were grouped into issues and issues into thematic areas. Votes could be delegated at three levels — globally, by area or by issue — and every vote and delegation was recorded and visible to members, trading ballot secrecy for the ability to hold delegates to account. The party's surge after its Berlin breakthrough in 2011, when it won 8.9% of the vote in the city-state election, made LiquidFeedback the largest delegative democracy platform in the world.

Participation

Participation was always a small fraction of membership. In August 2012, party officials reported about 10,000 registered users and estimated that 3–5% of roughly 30,000 members used the system actively. Kling, Kunegis, Hartmann, Strohmaier and Staab (2015) analysed complete daily dumps from 13 August 2010 to 25 November 2013 — 1,200 days — covering 499,009 votes on 6,517 initiatives in 3,565 issues, with 14,964 delegations. By January 2015 the installation had 13,836 registered users, against 24,438 party members. Activity rose with political events (the 2011 Berlin success, preparation for national programme conventions) and dropped after them. Voting activity itself followed a power law: 3,658 members voted more than 10 times, 1,156 more than 100 times and 54 more than 1,000 times.

Super-voters

The central finding concerns the distribution of voting weight. The weight per voter followed a power law with exponent 1.38: most voters held no delegations, while a small set held very large weights. Only 38 voters held more than 100 delegations. Delegates were more active than ordinary users, and delegations did not appear to follow shared political positions closely: delegates and their delegators voted identically 61% of the time before the delegation, against 51% for random pairs.

Did super-voters decide outcomes? Subtracting delegations from vote counts left 84.9% of outcomes unchanged; about one in six would have differed under direct voting. Measuring potential power (the ability to swing a vote) and exercised power (actually doing so), the authors found that the ability to decide votes grew roughly linearly with voting weight, but that super-voters used their power relatively less often than ordinary delegates — the average ratio of exercised to potential power was 0.34. Standard power indices (Shapley, Banzhaf) underestimated delegates' potential power. The authors emphasise that "though potential power might not be used at a given time, there is no reason to assume that this behaviour is stable."

Political trajectory

LiquidFeedback became a focus of internal conflict over whether its votes should bind the party, over anonymity and data protection, and over the influence of super-voters. The long-standing aim of turning the system into a binding "permanent members' assembly" (ständige Mitgliederversammlung) remained contested. The party's national fortunes declined: it won 2.2% of the second vote in the 2013 federal election, short of the 5% threshold. The LiquidFeedback case therefore shows a delegative system operating at scale inside a party for about three years, but not one that became the party's decision-making core.

Finding 2: LiquidFriesland — a public-sector pilot

The district of Friesland in Lower Saxony launched LiquidFriesland in November 2012, using LiquidFeedback to let residents propose and support initiatives for the district council. According to the Konrad Adenauer Foundation's summary of the district's own evaluation (dated 19 August 2015), 583 residents registered out of more than 80,000 eligible, and 382 used it at least once in nearly three years. Seventy-six initiatives were submitted by 30 different people. In the final twelve months only five initiatives went to a vote, none received more than 11 Yes votes, and the highest participation in any vote was 19. No activity was recorded after 10 April 2015. The platform cost about €7,000 a year; in the district council (Kreistag) it was described as "a very inexpensive PR exercise" for the district. The case is the clearest public-sector test of liquid democracy in Europe, and it failed to attract participants — not because delegation concentrated power, but because almost no one used it.

Finding 3: Google Votes, Democracy Earth and Decidim

Google Votes

Google ran an internal liquid-democracy experiment, Google Votes, on its corporate Google+ network. Its designers, Steve Hardt and Lia Lopes, described it in a 2015 technical disclosure that reports a case study of usage over three years, argues that social-networking technology can overcome the implementation complexity and infrastructure costs that had blocked adoption, and proposes a "Golden Rule of Liquid Democracy" for deciding how visible votes should be. The public abstract does not provide participation statistics comparable to the LiquidFeedback data.

Democracy Earth

The Democracy Earth Foundation's 2017 paper The Social Smart Contract proposed a blockchain-based liquid democracy in which a "vote" token confers participation rights. Its design includes basic delegation, delegation limited by topic tag, transitive delegation (which the delegator can switch off), the right to override a delegate's vote and public visibility of delegates' votes. To resist fake identities (Sybil attacks), it proposed a "Proof of Identity" process rewarding participants for simple tests designed to detect "replicants". The paper is a design document; it does not report results of use at scale.

Decidim

Decidim, the open-source participation platform first developed for Barcelona, does not offer transitive liquid democracy in its core. A community module, Action Delegator, adds three features to Decidim's elections component for cooperatives and similar organisations: voting restricted to a census, weighted votes by membership type, and delegation of a vote to another participant, with administrators able to limit the number of delegations per user and delegations logged for traceability. This is closer to traditional proxy voting at a general assembly than to liquid democracy, and it illustrates where delegation has found a stable niche: membership organisations that need a way for absent members to be represented.

Finding 4: DAOs — delegation at scale

DAOs on public blockchains are the largest current use of vote delegation. In protocols such as Compound, Uniswap and ENS, token holders must delegate their tokens — to themselves or to someone else — before they can vote, and delegates vote with the tokens delegated to them. Unlike most liquid-democracy designs, this delegation is one step only: delegates cannot pass votes on.

Fritsch, Müller and Wattenhofer (2022) analysed all holders, delegates, proposals and votes in the three systems from November 2019 to 1 March 2022:

On average, under 10% of total tokens (about 15% of circulating tokens) participated in votes, so 7.5% of circulating tokens would have sufficed to win an average vote; by comparison, participation at US shareholder meetings is around 70%. In Compound, around 20 delegates typically voted on a proposal. Many large delegates received most of their voting power from a single holder, raising the possibility that holder and delegate were the same entity. A follow-up study of 21 DAOs by Feichtinger, Fritsch, Vonlanthen and Wattenhofer (2023) found that in 17 of them a majority of voting power was controlled by fewer than 10 participants, that most voting power sat with delegates mainly representing a single token holder, and that governance transactions imposed significant monetary costs.

DAO delegation differs from civic liquid democracy because votes are weighted by token holdings rather than one person, one vote. But it shows the same dynamics at a larger scale: a few delegates accumulate decisive weight, and most holders neither vote nor delegate.

Finding 5: what theory says

Political philosophy

Blum and Zuber (2016) offer the most developed normative defence. Judged against representative democracy, they argue, liquid democracy mobilises more political expertise (on an epistemic account of democracy) and is more egalitarian (on an equality-based account). It faces two problems: it may recreate representative democracy's two classes of citizens through unequal voting power, and it may produce inconsistent policies across areas when different delegates decide related questions. They propose a modified model: delegates as trustees rather than mandated agents, a deliberative phase between delegation and voting so that "even an ordinary member with a single vote could bring forward the decisive argument that may convince a representative with 450 votes", and an executive that reviews feasibility and brokers package deals across policy areas. They leave the problem of secret voting for future work.

Computational social choice

Kahng, Mackenzie and Procaccia (2021) model a choice between a correct and an incorrect alternative and ask whether any delegation mechanism always does at least as well as direct voting, and sometimes better. Even assuming voters delegate only to better-informed voters, they show that local mechanisms — which use only each voter's neighbourhood and "arguably capture the spirit of liquid democracy" — cannot provide this guarantee, because concentration of votes in a few hands increases the variance of outcomes. A non-local mechanism, designed centrally, can outperform direct voting under mild assumptions. Gölz, Kahng, Mackenzie and Procaccia (2018) propose a partial remedy: let voters name several acceptable delegates and choose delegations so as to minimise the maximum weight any single voter holds.

Reading theory and evidence together

The empirical cases match the theoretical concern. LiquidFeedback's power-law distribution of weights and the DAO majorities held by a handful of delegates are precisely the concentration that makes local delegation risky. The empirical evidence also adds something theory underplays: most potential participants do not participate at all, so the question in practice is less whether delegation improves on direct voting than whether either mechanism attracts enough people to be legitimate.

A comparison of implementations

What the evidence does not show

• It does not show that liquid democracy produces better decisions. No study reviewed compares the quality of decisions under delegation with those under direct or representative voting in a real setting.
• It does not show how liquid democracy would work for a whole polity. All cases involve organisations with self-selected members, or a district pilot that few used.
• It does not show that concentration is inevitable. Designs that cap weights, require multiple delegates or decay delegations over time exist in theory but have barely been tested.
• It does not settle the secrecy problem. Public votes enable delegates to be checked but expose voters to pressure; LiquidFeedback chose transparency and no case has tested a secret-ballot alternative at scale.
• It does not show long-run behaviour of super-voters. Kling and colleagues found restraint over three years and warned that it need not last.

Implications for institution design

1. Cap or dampen accumulated weight. Limits on the number of delegations a person can hold (as in Decidim's module), multiple-delegate choices or weight decay address the super-voter problem directly.
2. Add deliberation between delegation and decision. Blum and Zuber's proposal — a reasoned debate that allows single voters to persuade heavy delegates — targets the main normative weakness.
3. Default to non-transitive or short chains. One-step delegation, as in DAOs, makes responsibility clearer, though it did not prevent concentration.
4. Decide on secrecy explicitly. Public delegate votes with secret ordinary votes is one compromise; it should be chosen, not inherited from software defaults.
5. Invest in participation before delegation. LiquidFriesland shows that the binding constraint is often whether anyone shows up, not how votes flow.
6. Use liquid voting where stakes and identities are manageable. Party opinion-forming, cooperatives and internal corporate choices have proved more suitable than public government, where identity verification, coercion and accountability are harder problems.

Open questions

• Can weight caps or multiple-delegate rules prevent concentration without discouraging delegation altogether?
• Would liquid democracy attract more participants if decisions were binding and consequential, or would it attract more capture?
• How can secret ballots be combined with accountable delegates?
• In DAOs, how much apparent delegation is actually large holders voting through their own delegate addresses?
• Could liquid delegation work as one component — for example within a legislature, or for agenda-setting — rather than as a replacement for elections?
• What would a well-designed public-sector pilot, with proper evaluation, look like?

Sources

1. Blum, C. & Zuber, C. I. (2016). Liquid democracy: potentials, problems, and perspectives. Journal of Political Philosophy 24(2), 162–182.
2. Kling, C. C., Kunegis, J., Hartmann, H., Strohmaier, M. & Staab, S. (2015). Voting behaviour and power in online democracy: a study of LiquidFeedback in Germany's Pirate Party. Proceedings of ICWSM 2015; arXiv:1503.07723.
3. Computerwoche (13 August 2012). Piratenpartei startet Demokratie-Software 2.0.
4. Bundeszentrale für politische Bildung (2013). Sonstige Parteien nach Zweitstimmenanteil, Bundestagswahl 2013.
5. Eisel, S. (2015). LiquidFriesland endgültig am Ende. Konrad-Adenauer-Stiftung, 3 September 2015.
6. Paulin, A. (2020). An overview of ten years of liquid democracy research. Proceedings of the 21st Annual International Conference on Digital Government Research, 116–121.
7. Hardt, S. & Lopes, L. C. R. (2015). Google Votes: a liquid democracy experiment on a corporate social network. Technical Disclosure Commons, Defensive Publications Series 79.
8. Democracy Earth Foundation (2017). The Social Smart Contract.
9. Decidim community. Decidim::ActionDelegator module (source code and documentation).
10. Fritsch, R., Müller, M. & Wattenhofer, R. (2022). Analyzing voting power in decentralized governance: who controls DAOs? arXiv:2204.01176.
11. Feichtinger, R., Fritsch, R., Vonlanthen, Y. & Wattenhofer, R. (2023). The hidden shortcomings of (D)AOs – an empirical study of on-chain governance. arXiv:2302.12125.
12. Kahng, A., Mackenzie, S. & Procaccia, A. D. (2021). Liquid democracy: an algorithmic perspective. Journal of Artificial Intelligence Research 70, 1223–1252.
13. Gölz, P., Kahng, A., Mackenzie, S. & Procaccia, A. D. (2018). The fluid mechanics of liquid democracy. arXiv:1808.01906; Web and Internet Economics (WINE 2018).