Broken feedback between elections: what officials hear when nobody is voting

In plain words

Broken feedback between elections is the problem that, in the years between votes, officials hear mostly from the people who show up, write in, or are organized to do so, and those voices can be unrepresentative or even fake. A report by the Office of the New York Attorney General found that nearly 18 million of the more than 22 million comments the Federal Communications Commission received in its 2017 net neutrality rulemaking were fake.

An election tells an official one thing: whether enough people preferred them over the alternative. It does not say which of their positions people liked, which they disliked, or what people want next. So between elections, officials listen to other signals: letters, calls, public hearings, comments on proposed rules, and polls. Each signal has a weakness. Hearings attract people with time and strong opinions. Comment systems can be flooded. Polls reach fewer people than they used to. The result can be that officials hear a distorted version of the public.

The problem, stated precisely

The problem has four parts:

1. Low information in elections. A vote for a candidate bundles many issues and cannot say which ones drove the choice.
2. Skewed participation in between. People who attend meetings or write to officials differ systematically from the public.
3. Manipulable channels. Open comment systems and contact campaigns can be flooded with fabricated or coordinated input.
4. Weak response loop. Even when input is representative, there is often no obligation for officials to respond to it or explain decisions.

This page does not cover elections themselves, which are covered on trust in elections and unequal participation. Coordinated campaigns as a measurement problem are on collective priorities.

Defined terms

• Feedback channel. Any route by which public preferences reach decision makers, including elections, public comments, meetings, contact, polls, and organized advocacy.
• Public comment. Written input submitted on a proposed rule or plan, often required by law before an agency acts.
• Thermostatic response. The pattern in which the public pushes back against the direction policy has moved. Bølstad describes it in elections as negative feedback from policy to election outcomes, in which a party that succeeds in setting policy tends to lose ground.⁠¹
• Sortition. Selecting participants by lot, used in citizens’ assemblies to produce a group that looks like the population.
• Response obligation. A rule requiring officials to say whether and how they will act on public input, and to justify rejection.

Evidence

Fake input in a federal rulemaking

The Office of the New York Attorney General investigated the Federal Communications Commission’s 2017 rulemaking on net neutrality and found that “fake comments accounted for nearly 18 million of the more than 22 million comments” the agency received.⁠² The fabrication came from both sides of the issue. The report found that a campaign funded by the country’s largest broadband companies generated more than 8.5 million fake comments to the FCC and more than half a million fake letters to Congress, using the names of real people without their consent, and that a 19-year-old college student who opposed repeal of net neutrality submitted more than 7.7 million fake comments using made-up identities.⁠² An open, count-based comment system produced a signal that was mostly noise.

Who speaks at local meetings

Local public meetings are a major feedback channel on land use and development. A study of meeting minutes in Massachusetts by researchers at the Boston University Initiative on Cities, as summarized by the Vermont Housing Finance Agency, found that homeowners were much more likely than renters to participate in housing-related meetings, and that participants were older, more likely to be male, and more likely to be white than their towns’ populations.⁠³ Of participants and commenters, 63 percent opposed housing initiatives and 14 percent supported them, while a statewide affordable housing referendum received 56 percent support from Massachusetts voters and won majorities in 65 percent of the communities studied.⁠³ The people officials heard from in the room held different views from the voters in their own communities.

Polls reach fewer people

Polls are a more representative channel by design, but response rates have fallen. Pew Research Center reported in 2019 that typical telephone survey response rates, which had held around 9 percent for several years, fell to 7 percent in 2017 and 6 percent in 2018.⁠⁴ Low response does not automatically mean bias, but it makes accuracy depend more heavily on weighting and on who chooses to answer; see measurement.

Elections send a coarse signal

Research on thermostatic opinion suggests the public does respond to policy between elections, but in a coarse way. Jørgen Bølstad’s 2012 study in PS: Political Science and Politics reports evidence of “thermostatic voting in American presidential elections,” in which a party that succeeds in moving policy tends to lose electoral ground.⁠¹ That is feedback, but it arrives years later and says only “too much” or “too little” in the aggregate, not which policy or by how much.

Hypothetical example: three signals, three answers

Hypothetical example. A city council considers a new bus lane. Forty residents speak at a hearing, thirty of them against. The council’s inbox holds 2,000 identical emails in favor, sent through an advocacy group’s form. A probability-based poll of 600 residents finds a narrow majority in favor, with a margin of error of several points. Each signal is real, and each answers a different question: who cares enough to attend, who can be organized to click, and what a representative sample says when asked. A council that counts only the hearing, only the inbox, or only the poll will reach different conclusions. Good feedback design does not pick one signal; it labels each for what it is.

Causes and competing explanations

• Bundling. Elections combine many issues into one choice of candidate or party.
• Participation costs. Attending a weekday evening meeting or writing a formal comment takes time, knowledge, and confidence, which are unevenly distributed.⁠³
• Open systems invite manipulation. Comment portals that count submissions without verifying them reward volume, which can be manufactured.⁠²
• Declining survey participation. Fewer people answer polls, raising cost and uncertainty.⁠⁴
• No duty to respond. Officials often receive input without any obligation to explain what they did with it, which discourages participation.
• Competing explanation: representation by design. Some argue that officials are elected to use judgment between elections, not to track every shift in opinion, and that a coarse signal is appropriate; see representation.

Who is harmed

People without time or resources to attend meetings, renters in land-use decisions, and groups underrepresented among commenters are harmed when skewed channels drive decisions.⁠³ Real commenters are harmed when their voices are buried among fabricated ones, and people whose identities were used without consent are harmed directly.⁠² Officials are harmed in their ability to know what their constituents think. And the public is harmed when agencies or legislators can point to manufactured volume as evidence of public support.

What has been tried elsewhere

Permanent citizen deliberation linked to parliament. The German-speaking Community of Belgium has institutionalized a permanent citizens’ dialogue with three parts selected by lot: a permanent Citizens’ Council that chooses topics, Citizens’ Assemblies that deliberate and make recommendations, and a joint committee where assembly members, elected representatives, and the responsible minister discuss them. Representatives and the minister must say whether and how they will implement each recommendation, and rejections require specific justification.⁠⁵ Results: it builds a response obligation into the process. Limits: published outcomes evidence was not available in the abstract reviewed.

Online consultation with consensus mapping. vTaiwan, begun in 2014, uses stages of proposal, opinion, reflection, and legislation; its organizers report that more than 28 cases have been discussed and that about 80 percent led to decisive government action, including a financial technology sandbox law.⁠⁶ These are the organizers’ own figures.

City participation portals. Reykjavík offers a consultation portal for projects in progress, a suggestion portal, youth councils, citizen conventions, and a neighborhood program in which residents propose and vote on district projects funded with 850 million krónur every other year.⁠⁷

Verifying comment identity. The New York Attorney General’s findings point toward verification and disclosure as responses to comment fraud.⁠² Limits: identity checks can deter legitimate commenters and raise privacy concerns.

Representative polling. Probability-based polls remain the standard for representative opinion, despite falling response.⁠⁴

Competing interpretations

The more-participation view holds that officials need richer and more frequent public input, and that permanent, representative bodies like the Belgian model,⁠⁵ and digital consultation like vTaiwan⁠⁶ can supply it.

The better-participation view holds that more of the same channels would amplify their skew⁠³ and that what matters is who participates and how input is weighed. Supporters favor sortition and probability sampling over open channels.

The trustee view holds that elected officials are chosen to exercise judgment, that the public sends meaningful feedback through elections even if coarse,⁠¹ and that constant consultation can make officials followers of the loudest voices. On this view, the problem is overstated.

The integrity view holds that the first priority is preventing manufactured input, since fake volume on both sides of a single proceeding shows the risk is not partisan.⁠²

The privacy view warns that fixing feedback by verifying every participant’s identity can create records of who said what, which can chill participation, especially on sensitive subjects.

Application to this framework

My Vote Forward is designed around this problem. Its participation over time chapter describes participation between elections, from pre-proposal through follow-through. Priorities separate what matters most from specific opinions, so officials can see both without one being mistaken for the other. Published releases are scheduled and reviewed, not live counters that can be flooded. The representation chapter covers explanation of changed positions, which is the response half of feedback. The collective priorities chapter covers coordinated participation and why aggregates carry no automatic mandate.

Unresolved questions

• Can an open feedback channel be protected from fabrication without identity requirements that chill participation,⁠²?
• Do response obligations, like the Belgian model’s⁠⁵ change what officials do, or mainly what they say?
• How should officials weigh representative but small samples against large, self-selected, and intensely felt input?
• What frequency of feedback helps governance, and when does it become noise?