In “Rank, Pack, or Approve,” Lodewijk Gelauff and Ashish Goel introduce a dataset drawn from the novel and comprehensive Stanford Participatory Budgeting platform. The data span over 150 real participatory budgeting processes, or “elections.” Importantly, the elections vary in terms of how ballots are designed and how participants make budgeting decisions. Gelauff and Goel ask how such variation shapes important budgeting outcomes, such as when participants are more likely to become fatigued and abandon the process.
Two key findings from the study are as follows: First, more complex PB designs lead voters to, perhaps unsurprisingly, spend more time participating; however, this does not significantly increase abandonment or “dropout” rates. Second, voting methods that force participants to deal with cost trade-offs — as opposed to merely indicating their preferences — have been found to generate less expensive projects.
The reader comes away with a sense of how subtle differences in the design of budgeting elections meaningfully shape the allocation of resources. This will resonate with social scientists who are familiar with how, for example, different kinds of electoral rules shape political competition. To understand Gelauff and Goel’s findings, it helps to first outline how PB elections differ from one another.
Ballot Design and Voting Methods:
The basic PB setup involves organizers choosing a voting method, a list of projects to potentially be funded, and an authentication process (i.e., checking that participants are valid voters). Voters then select or rank projects given the constraints of each voting rule or method. These three rules, captured in the paper’s title, are as follows:
The first, “K-approval,” asks voters to select up to “K” projects. The top-voted projects receive funding until the budget runs out. K-approval is simple, but its main drawback is that it ignores the costliness of each project: voters only indicate which projects they like, rather than how those choices fit within a fixed budget. The second method, “K-ranking,” asks voters to rank their preferred projects, capturing their preferences in a more fine-grained manner. As votes are aggregated using the Borda scoring method, higher-ranked projects receive greater weight or value. Finally, the “knapsack” method asks voters to choose projects that fit within a fixed budget. This method best allows participants to balance costs in a way that mimics real city councils. However, knapsack is more complex and time-consuming than K-approval or K-ranking, although the online interface design, which mimics a shopping cart, is already much simpler than it would be on paper.
Data Collection and Findings:
As mentioned, Gelauff and Goel’s data is drawn from the open-source Stanford PB platform. This tool enables cities to conduct online PB elections with a great deal of customizability, including location, budget, language of operation, authentication process (e.g., requiring personal information or sending SMS messages), as well as methods, phases, and windows of voting. Key for the authors’ purposes, it also tracks (anonymous) voters’ choices and how much time they spend during the election. Data collection began in 2014.
The first key finding is motivated by the fact that election organizers often prefer K-approval for its simplicity. As such, Gelauff and Goel analyze how much time participants spend on their ballots and how often they quit. Although more complex ballots — those with a larger budget and number of projects — are shown to predict longer completion times, they do not significantly increase dropout rates. The authors note that more research is needed to assess whether knapsack specifically affects dropout.
The authors also find that voters select more expensive projects with K-approval compared to the knapsack methods. However, voters indicate similarly expensive preferences for their most-preferred projects under both methods; the key difference appears lower down the list of preferences, where the knapsack constraint forces them to be more cost-conscious. In other words, the knapsack cost constraint doesn’t affect which expensive project participants most prefer. Rather, it limits how many extra expensive projects they can add.
Overall, “Rank, Pack, or Approve” deepens our understanding of how PB can improve direct democratic engagement while reducing burdens on participants. It does this while providing a large quantity of real-world data, compared with prior research that has relied on crowdworkers without a real stake in the budgeting outcome. The authors helpfully illustrate how local governments can design PB processes that are clearer and more inviting to ordinary voters. Subsequent research will benefit from using this powerful data resource, as will organizers seeking to expand local engagement.
*Brief prepared by Adam Fefer.
