Referendum failures: when asking everyone produces an answer nobody can use

In plain words

A referendum failure is a public vote that produces a result people cannot clearly act on, do not accept as fair, or regret because the question hid what it would actually do. Referendum failures are usually caused by how the vote was designed, not by voters, and international guidance from the Council of Europe’s Venice Commission sets out design rules meant to prevent the most common ones.

Here is the simple version. A referendum asks millions of people one question with two answers. That sounds clear, but the real decision usually has many parts. What happens next? Who carries it out? What if the answer turns out to be impossible? If the question does not say, people may vote for different things while marking the same box. Some votes also have rules about how many people must show up, and those rules can let a small group win by staying home.

The problem, stated precisely

A referendum fails, for this page’s purposes, when at least one of these happens:

1. The meaning of the result is unclear. Voters chose an outcome without a defined path to it, so the winning side disagrees with itself about what it won.
2. The rules distort the result. A turnout or approval quorum turns abstention into a tactic, or a threshold leaves a clear majority without a valid result.
3. The question bundles unrelated choices. Voters must accept or reject several provisions at once.
4. The result locks in a rule that later fails. A binding measure, often a budget formula, cannot adapt when circumstances change.
5. The legitimacy of a narrow result is contested. A very close or low-turnout vote settles a large question that a later government or court must still implement.

This page does not argue that referendums are bad. Many are clear, accepted, and carried out. It also does not repeat material covered elsewhere: the risk that a majority votes away a minority’s rights is on tyranny of the majority; the legal status of referendums and the history of consultation are on participation between elections; and the role of money in initiative campaigns is on money in politics.

Defined terms

• Referendum. A vote of the electorate on a proposal rather than on a candidate. It may be binding or advisory, and it may be required by law, called by an authority, or triggered by citizens.
• Initiative. A process by which citizens collect signatures to place a proposal on the ballot.
• Turnout quorum. A rule that a referendum is valid only if a minimum share of registered voters takes part.
• Approval quorum. A rule that a proposal passes only if the yes votes equal a minimum share of all registered voters, not just of those voting.
• Unity of content. The principle that the parts of a single question must have an “intrinsic connection” so voters are not asked to accept or reject unrelated provisions as a block.⁠¹
• Information shortcut. A cue, such as which groups support or oppose a measure, that lets a voter make a reasonable choice without studying every detail.⁠²

Evidence

International design standards

The Venice Commission’s Code of Good Practice on Referendums, adopted in 2006 and 2007, is the most detailed international benchmark. It is advisory, not binding law. It states that “The question put to the vote must be clear; it must not be misleading; it must not suggest an answer; electors must be informed of the effects of the referendum”.⁠¹ It says authorities “must provide objective information,” including an explanatory report that gives “a balanced presentation” of supporters’ and opponents’ views.⁠¹ On quorums, it says “It is advisable not to provide for” a turnout quorum, “because it assimilates voters who abstain to those who vote no,” or an approval quorum, which “risks involving a difficult political situation” when a measure wins a majority below the threshold.⁠¹

The Code’s explanatory report gives a worked example of the turnout-quorum problem: if 48 percent of electors favor a proposal, 5 percent oppose it, and 47 percent intend to abstain, “the 5% of opponents need only desert the ballot box in order to impose their viewpoint”.⁠¹

Failure by quorum: Hungary, 2016

On October 2, 2016, Hungary held a referendum on European Union migrant relocation quotas. According to the National Election Office as reported by Al Jazeera, turnout was 43.9 percent, below the 50 percent needed for validity, while 98.3 percent of valid ballots backed the government’s position.⁠³ The vote was declared invalid. The prime minister nonetheless called the result “an outstanding result” and proposed putting it into the constitution.⁠³ Both sides could claim the outcome: opponents, many of whom abstained, because it was invalid, and the government, because of the lopsided yes share. This is the ambiguity the Venice Commission’s quorum guidance warns about.

Failure by narrow margin on a large question: Colombia, 2016

On October 2, 2016, Colombian voters were asked: “Do you support the final accord to end the conflict and build a stable and lasting peace?” According to IFES ElectionGuide, No received 6,431,376 votes (50.21 percent) and Yes 6,377,482 (49.78 percent).⁠⁴ ElectionGuide lists 13,066,047 votes cast and 34,899,945 registered voters, which works out to turnout of roughly 37 percent (our calculation from those two figures). A peace agreement negotiated over years was rejected by a margin of about 54,000 votes, with most registered voters not voting. What should follow such a result is a political question the ballot itself could not answer; the dedicated case page on the agreement’s later history is part of the cases library.

Failure of meaning: the United Kingdom, 2016

In the June 23, 2016 referendum on European Union membership, Leave received 17,410,742 votes (51.89 percent) and Remain 16,141,241 (48.11 percent), with turnout of 72.21 percent, according to figures reported from the Electoral Commission; Scotland voted 62.00 percent Remain and Northern Ireland 55.78 percent Remain.⁠⁵ The question asked whether to remain or leave, not what relationship should replace membership. The legal effect of the result, and the Supreme Court’s 2017 ruling that implementing it required Parliament, are covered on participation between elections. For this page, the point is narrower: a binary question about a many-sided decision left the winning majority without an agreed definition of what it had chosen.

Failure by lock-in: California’s Proposition 98

California voters approved Proposition 98 in November 1988 with 51 percent, setting a minimum annual funding level for K to 14 education.⁠⁶ The state’s nonpartisan Legislative Analyst’s Office reported in 2017 that “the minimum guarantee is now governed by eight interacting formulas and nearly a dozen different inputs,” that the formulas “have muddled the budget process,” and that it found “no clear evidence that school funding is higher today” because of them.⁠⁶ Supporters of the guarantee see it as protection for schools against legislative raids; the Analyst’s finding is that the design is hard to understand and adapt. The wider budget issue is on public budget blindness.

Low turnout and rejection after deliberation: Ireland, 2024

Ireland’s experience shows that a careful process before a vote does not guarantee approval. In March 2024, voters rejected two proposed constitutional changes on family and care. The care amendment was rejected 73.9 percent to 26.1 percent, the highest No share in an Irish referendum, and the family amendment 67.7 percent to 32.3 percent, with turnout around 44 percent.⁠⁷ The care proposal followed a recommendation from the Citizens’ Assembly on Gender Equality to replace Article 41.2 with gender-neutral text supporting care in the home and community.⁠⁸ Whether the government’s final wording matched the Assembly’s intent is a question this page cannot settle from the sources reviewed.

Causes and competing explanations

• Compression. A multi-part decision is reduced to two answers. Every referendum does this; the failure comes when the parts that were dropped are the ones that matter most.
• Agenda control. Whoever drafts the question decides what is compressed. Governments sometimes call referendums for tactical reasons, and initiative sponsors write the text they want. See agenda capture.
• Rule design. Quorums, thresholds, and bundling are chosen by legislators and can be set to make a given outcome more or less likely.⁠¹
• Information. Critics argue voters cannot master complex measures. Arthur Lupia’s 1994 study of five complicated California insurance initiatives found that “access to a particular class of widely available information shortcuts allowed badly informed voters to emulate the behavior of relatively well informed voters”.⁠² The competing explanation, then, is that information is not the main problem when good cues exist, and becomes a problem when cues are absent or misleading. See misinformation.
• Turnout. When large shares do not vote, a result can be decisive in law and weak in legitimacy, which feeds disputes about what “the people” decided. See unequal participation.

Who is harmed

The first group harmed is voters themselves, who are asked to settle a question whose consequences were not explained and who may later be told their choice meant something they did not intend. The second is the people affected by an outcome that is hard to implement or revise, such as students and taxpayers under a budget formula that its own analysts describe as muddled.⁠⁶ The third is the institution of the referendum: each failure gives critics evidence that direct votes are unreliable, which can make it harder to use the device well. Minorities face the specific risks discussed on tyranny of the majority.

What has been tried elsewhere

Clear-question and balanced-information rules. The Venice Commission’s Code sets these as good practice.⁠¹ Results: the Code gives courts, observers, and legislators a common standard. Limits: it is advisory, and it cannot stop a government from holding a vote on a vague question if domestic law allows it.

Avoiding quorums. The Code recommends no turnout or approval quorums.⁠¹ Hungary’s 2016 vote shows what can happen with one.⁠³ Critics of this advice argue that some threshold protects against small, motivated electorates deciding constitutional questions; the Code’s answer is that abstention should not be counted as a vote.

Citizen review panels. Oregon’s Citizens’ Initiative Review convenes randomly selected voters to study a ballot measure and write a statement printed in the official Voters’ Pamphlet. An evaluation by Knobloch, Gastil, Richards, and Feller of the 2012 reviews found the process “highly deliberative,” found the statements had “no obvious factual errors or misleading sentences,” reported that 51 percent of likely voters were aware of the review by the end of the 2012 election, and found “substantial knowledge gains” in an online experiment for one measure.⁠⁹ Limits: the evaluators were academic researchers studying a program they had helped assess since 2010, awareness was about half of likely voters, and knowledge gains were measured for one measure.

Citizens’ assemblies before a referendum. Ireland’s 2016 to 2018 Citizens’ Assembly considered the Eighth Amendment among other topics,⁠¹⁰ and the 2018 referendum on that amendment passed 66.40 percent to 33.60 percent, with roughly 64 percent turnout by our calculation from ElectionGuide’s figures.⁠¹¹ The 2024 care and family votes, which also followed assembly and parliamentary committee recommendations, were rejected by wide margins.⁠^7,8 Deliberation can shape the question; it does not decide the vote.

Information shortcuts. Rather than trying to make every voter an expert, some argue for making cues reliable: disclosure of who funds each side, official neutral summaries, and endorsements from known groups. Lupia’s findings support the idea that good cues help.⁠²

Competing interpretations

Defenders of direct democracy argue that referendum “failures” are often outcomes elites disliked. On this view, the Hungarian, Colombian, and British votes expressed real public sentiment, and calling them failures disrespects voters. They point to research such as Lupia’s showing that voters can use cues effectively,⁠² and to Switzerland’s long, routine use of direct votes (see participation between elections).

Representative-government advocates argue that complex, many-sided decisions belong with elected bodies that can negotiate, amend, and be held accountable at the next election, and that a binary vote cannot carry instructions for implementation. Proposition 98’s formulas are an example of a decision fixed in place that later needed more flexibility than a ballot measure allows.⁠⁶

Reformers in between argue the device is sound but needs guardrails: clear questions, no quorums, unity of content, neutral information, and pre-vote deliberation.⁠^1,9 Critics of reform warn that every guardrail gives someone, such as a commission or a court, power to shape what voters see.

Skeptics of turnout thresholds and skeptics of their absence both have a point. Without a threshold, a small electorate can change a constitution. With one, abstention becomes a weapon.⁠^1,3 There is no rule that avoids both risks.

Application to this framework

My Vote Forward’s proposals chapter separates an issue, a proposal, and a ballot item, and asks who has authority, what it costs, and what depends on it. That separation is aimed at the compression problem: before something is a yes or no question, its parts should be visible. The participation over time chapter describes stages before and after a vote, so a result is not treated as the end of the story. The question design chapter covers wording and order effects that apply to referendum questions as much as to surveys. And private voting keeps the official decision with the official ballot, under the law of the jurisdiction.

Unresolved questions

• Is there a quorum rule that protects against small electorates deciding constitutional questions without turning abstention into a tactic?
• How much detail should a referendum question carry about implementation before it becomes too long for voters to read?
• When an advisory referendum produces a narrow or low-turnout result, what obligations do elected officials have? See representation.
• Do citizen review statements change outcomes, or only knowledge? The 2012 Oregon evaluation measured awareness and knowledge, not vote shifts.⁠⁹
• Should binding budget formulas adopted by initiative carry automatic review dates?