Concentration of Power and Participation in Online Governance: the Ecosystem of Decentralized Autonomous Organizations

Blockchain technology enables a new form of online community: Decentralized Autonomous Organizations (DAOs), where members typically vote on proposals using tokens. Enthusiasts claim DAOs provide new opportunities for openness, horizontality, and democratization. However, this phenomenon is still under research, especially given the lack of quantitative studies. This paper presents the first census-like quantitative analysis of the whole ecosystem of DAOs, including 30K DAO communities on the main DAO platforms. This enables us to provide insights into the allegedly "democratic'' nature of DAOs, building metrics concerning their lifespan, participation, and power concentration. Most DAOs have a short lifespan and low participation. There is also a positive correlation between community size and voting power concentration. Like other online communities, DAOs seem to follow the iron law: becoming increasingly oligarchic as they grow. Still, a significant amount of DAOs of varying sizes defy this idea by being egalitarian by design.


INTRODUCTION
Recently, we have observed the emergence of the so-called "Web3" with a new wave of decentralized technologies (e.g., blockchain, IPFS) that enable a new form of self-governed online communities: Decentralized Autonomous Organizations (DAOs).
DAOs allow individuals to self-govern and coordinate through a set of self-executing rules deployed on a public blockchain, where governance is independent of central control [4].DAO members use "governance tokens" to participate in the DAO decision-making process, often through a voting system.These tokens can represent monetary value (e.g., cryptocurrencies), voting power, permissions, and/or reputation.DAOs typically manage crypto-assets, such as cryptocurrencies, and their members may propose how to allocate these assets through proposals that align with the collective interest.
Web3 enthusiasts claim the advent of DAOs provides a new opportunity for openness, horizontality, and, in sum, democratization [10].DAOs indeed provide open participation, transparent economics, or decentralized infrastructure, and many try to reflect egalitarian ideals.And yet, it is unclear to what extent these features have enabled truly democratic self-governed communities or organizations more akin to elite-driven oligarchic plutocracies.This paper explores this question by studying the full picture of the DAO ecosystem.
Today, DAOs are an undeniable reality, with thousands of DAO communities jointly managing $29B (see deepdao.io).However, there is a lack of large-scale quantitative works that provide insights into this reality.Most of the literature focuses on single case studies (e.g.[6,13]), or small subsets of DAOs [8].The few existing larger quantitative studies [3,11] do not explore power concentration, nor do they cover the whole ecosystem of DAOs.
This work provides the first census-like study of the entire DAO ecosystem.This enables us to explore critical questions concerning the democratic features of DAOs.For example, as peer production communities grow, power concentration typically increases [9], which makes it harder to consider them democracies.On another matter, the inequality of participation is a common property in online communities [7,12]

DATA RETRIEVAL AND PREPARATION
In this section, we describe the creation of our dataset.The resulting dataset, with over 10K DAOs and nearly 22M votes, aims to provide a comprehensive view of decision-making within DAOs and is made public to foster further research on DAO governance 1 .
Note that our units of analysis are "DAO deployments" in the blockchain 2 .In principle, a DAO can have several deployments, either because they are used for different purposes or because they have been replaced by a newer one, but it is not possible to track such aspects automatically.Thus, we will analyze DAO deployments, but will refer to them as DAOs for clarity.

Data retrieval and cleaning
The vast majority of DAOs are deployed using a DAO platform.These software platforms facilitate the creation, deployment, and use of DAOs.Our analysis includes the DAO governance platforms listed in DeepDAO, a leading website for DAO analytics.These platforms include Aragon, Colony, DAOHaus, DAOstack, OpenLaw, Realms, Snapshot, Tally, and Substrate.
Data was collected from July 20 to August 8, 2023.Our data retrieval method consisted of using the platform's API to retrieve data about the DAOs deployed by the platform 3 .Thus, we excluded those platforms that provide no API, namely OpenLaw, Colony, and Substrate, which accounted for a few dozen DAOs according to DeepDAO.
Next, we developed a universal schema for votes, proposals, and DAOs, which included information about the number of proposals and the votes cast in each of them.As DAOs are transparent platforms, the vote information identifies the voter and their voting power in that vote.As a result, we obtained a dataset with 30,250 DAO deployments, 207,131 proposals, and 21,619,733 votes.
In the retrieved data we observed votes with negative or null weights (less than 0.015%).We assumed it corresponded to errors and removed them from the dataset.More importantly, we also discovered multiple deployments with no activity, perhaps created by mistake or for casual testing.Thus, we excluded DAO deployments without proposals and those with no more than one distinct voter.After the data cleaning, the dataset accounts for 10,541 DAO deployments with 186,228 proposals and 21,606,463 votes. 1 The data is publicly available at Zenodo https://doi.org/10.5281/zenodo.10794916. 2Including Ethereum mainnet, Ethereum-compatible chains, and Solana among others. 3Some platforms lacked official data retrieval methods.In these cases, we collected information through the undocumented APIs that power the public-facing websites and The Graph, a decentralized protocol for indexing and querying blockchain data.

Voters and voting power distribution
To estimate the number of voters for each DAO, we counted the unique number of blockchain addresses involved in the DAO's voting process.There is no guarantee that an address corresponds to a unique voter, or that each voter has only one address.This is a typical problem in online communities.For the count, we refer to unique voters to denote unique addresses.Our dataset includes 5,052,404 unique voters, with each voter being considered only once, even if they participated in multiple DAO deployments.
Regarding the distribution of the voting power among DAO members, such information is not provided by the APIs.Voting power typically depends on token holdings of one or several kinds and/or on the voting system used by the DAO.All the APIs provide the voting power of the votes cast by voters on each proposal.Still, each proposal is typically voted on by only a few DAO members, and the voting power distribution usually changes over time as members can join or leave the organization and their voting power may also change over time.Furthermore, some DAO members may have never voted.
Taking all these considerations into account, we propose a way to approximate the voting power distribution of each DAO by considering the median voting power of each voter's votes within a DAO.The median voting power gives us a conservative estimation of each voter's voting power that is not affected by outlying values.As a result, we can create a conservative estimation of the voting power distribution.We assume that most DAOs undergo minimal turnover in their membership and that for most members their voting power does not greatly change.
The metric only considers the voting power of individuals who have participated at least once in the voting process, with the understanding that those who have never participated do not contribute to the DAO's estimated voting power.Hence, the resulting estimated voting power primarily reflects the effective voting power employed in governance.With this metric, we acknowledge a certain level of noise, but it gives an operational estimation of a dynamic element like the voting power.is probably because it made off-chain 4 voting possible, which provides better usability than on-chain voting and avoids blockchain transaction costs.The other platforms are significantly behind Snapshot, including pioneering platforms such as Aragon or DAOstack, which have been operating since 2018 and 2019, respectively 5 .

FINDINGS AND INSIGHTS 3.1 Organization size and period of activity
If we look at the DAO sizes, we can see that small organizations are prevalent: 50% of the DAOs have 10 or fewer voters.As the size of an organization increases, there are fewer DAOs.Still, the number of organizations with more than 1K voters is close to 300, and 4 of them have more than 100K voters.This means that it is possible to find DAOs governed by extremely large communities, but the vast majority have fewer than 100 voters.
Regarding the lifespan or activity period of DAOs, in Table 2 we measure the activity period of a DAO as the number of days between its first and last proposal up to the date of analysis.We find that, in general, the bigger the DAO (in terms of number of voters), the older it is.This is to be expected because organizations grow over time, and large organizations are usually the result of sustained activity.However, there are exceptions, such as the only DAO with more than 1M voters, which was active for only 43 days.
Roughly speaking, we considered a DAO to be active if it had submitted a proposal in the last 6 months.According to this metric, around 53% of the DAOs were active at the time the dataset was created.Even with the understanding that there is some noise in the metric, the data suggests that about half of the DAOs in the census are not active.The percentage of inactive DAOs decreases with the DAO size, which may suggest that as a DAO grows in size, it is less likely to perish (or to abandon a given deployment), and vice versa.

Concentration of power and participation
We analyze the voting activity across DAO size using Table 3 6 .Regarding the number of proposals, the median values reveal that at least half of the DAOs exhibit moderate behavior regardless of their size.For example, the median number of proposals put forward in DAOs with between 1K and 10K voters is 20.Still, we can also find DAOs that exhibit extremely active behavior (e.g., the maximum number of proposals in a single DAO is around 29K).
Regarding the percentage of proposals a voter participates in, voters tend to be more involved in small DAOs, as the number generally decreases with the DAO size.In bigger DAOs, the central tendency values reveal that most voters vote in less than 30% of the proposals.Similarly, as the size of the DAO increases, the voter turnout decreases.Thus, proposals typically have lower participation rates in bigger DAOs.While this result is not surprising, the highest median value is very low, with just one-third of active voters in DAOs with 2 to 10 voters.Moreover, the values drop sharply, revealing that in bigger DAOs many voters hardly participate7 .Regarding the distribution of voting power, the Gini index shows that as the size of the DAO increases, so does the level of inequality, i.e. power is more concentrated in the hands of few voters.To validate this relationship, we computed the correlation between the Gini index and the DAO size as the number of voters of DAOs in decimal logarithm.Pearson's correlation coefficient resulted in r = 0.4819, and Spearman's rank correlation coefficient is  = 0.4818, both of them significant at the 0.01 level of significance.These values confirm a direct relationship between the size of a DAO and the concentration of power, both linearly (Pearson) and monotonically (Spearman).The scatter plot in Figure 1 confirms this behavior but also reveals an interesting departure from it.In particular, at the bottom of the figure, we can appreciate a high number of DAOs of various sizes with zero Gini (39% of the total), that is, DAOs with a one-person-one-vote voting system.Of these, 22 DAOs (0.5% of the total) have more than 1K voters.Thus, many DAOs seem to be governed by egalitarian principles, and, among them, a few of them are quite large in terms of number of voters.
We also created a metric to assess the inequality of the voting power distribution in a more meaningful way than the Gini index.In particular, we compute the minimum percentage of voters needed to reach more than 50% of the voting power of the DAO.This metric decreases sharply with the size of the DAO.In the majority of DAOs with more than 1k voters, more than 50% of the voting power is controlled by an oligarchy of around 1% of the voters or less.This fact reveals that most of the larger DAOs, even if they are  3: Voting activity metrics.For metrics with (*), we first compute the mean value for each DAO, and then we compute the summary statistic for the DAOs within each size category.democratic or even horizontal, are very far from being egalitarian organizations.

CONCLUDING REMARKS
This work presents the first census covering all currently relevant DAO platforms, resulting in 10K DAOs.The dataset is open-licensed and available for further research.The analysis conducted enables us to extract several characteristics of this new type of community on a large scale and draw parallels with other online communities.
First, 20K of the initial 30K DAOs were empty DAOs, likely used for testing.As with other software tools, casual or testing use is common, but on the blockchain, it is permanently recorded and may distort the perception.More importantly, numerous DAOs are inactive or abandoned.This may suggest that, given the Web3 hype, many people started DAO projects but did not continue their endeavors.Failure is a pattern well-documented in other online collective projects, such as FLOSS and wiki projects [1].
Second, we can see how the majority of DAOs exhibit a small to moderate size (in terms of unique voters), while a few stand out as significantly larger.Again, this duality is reminiscent of patterns observed in wiki-based and open-source communities [2].
Our analysis also reveals remarkable voting patterns.As DAO size increases, its voter turnout and the number of proposals a voter participates in greatly decrease.While the presence of passive members is also common in online communities [5], it poses a challenge to the scalability of representative governance in DAOs.
We also observe high levels of inequality in the distribution of voting power.Notably, there is a statistically significant correlation between larger DAOs and greater inequality.This prompts us to consider inequality as a typical "feature" of collaborative online communities once they reach a certain size, suggesting that DAOs may conform to the "iron law of oligarchy" [9].This is particularly relevant for DAOs, since "power can be bought" quite literally, as many DAOs allow the purchase of governance tokens.Thus, mitigating power imbalances within these communities may be a critical consideration for fostering more inclusive and equitable decision-making processes.However, we do identify a small number of large egalitarian DAOs that defy the iron law of oligarchy and demonstrate the variety of ways in which these organizations can present themselves.
In sum, through exhaustive data collection and in-depth quantitative analysis, this work brings valuable insights into the power distribution and participation patterns within the DAO ecosystem.This work, along with its public dataset, serves as a base to delve into new research questions about these communities, such as how inequality affects governance effectiveness or decision-making outcomes.

Figure 1 :
Figure 1: Scatter plot of the Gini index versus the DAO size.

Table 1 :
. Do DAOs exhibit such unequal and oligarchic patterns or behave in a more egalitarian manner?Distribution of DAOs across platforms and across DAO size in number of voters.

Table 1
shows the distribution of DAOs by platform and size, measured as the number of unique voters.Snapshot stands out as the most popular platform with more than 8K DAOs (over 80% of the total) and also as the platform hosting the largest DAOs.Its popularity

Table 2 :
Metrics of the period of activity of the DAOs.