Referendum vs. Plebiscite: The Difference and Why It Matters

The words referendum and plebiscite get used interchangeably, and most of the time the confusion is harmless. It stops being harmless the moment someone has to draft the authorising statute, write the ballot question, or plan what happens after the count. In operational terms the two instruments differ on four testable parameters: who initiates the vote, what the ballot question attaches to, what threshold arithmetic governs the result, and what the implementation path is if yes wins. Get those four right and the label printed on the ballot is a local naming convention. Get them wrong and you inherit a design flaw with a publish date: voided results, contested thresholds, and litigation over legal effect that starts after the votes are counted.

A referendum is a vote in which the electorate decides on a specified legal text — a bill, a constitutional amendment, a treaty — under rules that fix the vote’s legal effect before the ballot is printed. A plebiscite is a vote in which the electorate expresses approval or preference on a political question whose legal consequence is determined after the result, by the authority that called the vote. The binding-versus-consultative shorthand is real, but it is a property of the statute, not of the word at the top of the ballot.

Read the Act, Not the Label

Jurisdictions do not agree on vocabulary, and the disagreement is itself a design hazard. France uses référendum for votes under Articles 11 and 89 of its constitution, including votes that function politically as confidence tests. The United Kingdom’s 2016 vote on EU membership was labelled a referendum, but the European Union Referendum Act 2015 said nothing about legal effect; the vote was advisory in law, and implementation was settled by Parliament afterwards. Colombia’s 2016 vote on the FARC accord was labelled a plebiscite and was non-binding by statute. Switzerland labels everything a votation and distinguishes instruments by signature rules and subject matter instead. The classification method on this site therefore ignores labels entirely: read the authorising act and ask what it fixes. The four parameters below are the working version of that method.

Parameter 1: Initiation and the Verification Chain

Referendums are usually initiated by rule: a citizen petition with a signature threshold, a legislature referring its own text, or a constitutional requirement that certain measures go to the voters. Switzerland’s federal popular initiative requires 100,000 signatures within 18 months (Article 139 of the Federal Constitution); the optional referendum against a statute requires 50,000 signatures within 100 days (Article 141). Signatures are certified by the communes, validated federally, and published on a fixed timetable. That is a verification chain: named actors, defined checks, deadlines. In systems with large petitions, verification becomes a subsystem of its own. California permits random-sample verification for initiative petitions (Elections Code § 10504), trading audit cost for a stated confidence level, and the Venice Commission’s Code of Good Practice on Referendums accepts sample-based signature checks for the same reason: a full check of 100,000 signatures is rarely proportionate. The design rules behind those checks are covered in signature verification rules.

Plebiscites are initiated by decision: a cabinet resolution, a presidential decree, a parliamentary motion. There is no petition to verify because there is no petition. The design consequence is that the only quality controls left in the system exist on election day, and every upstream choice — timing, question wording, franchise — arrives at the count unfiltered.

Parameter 2: Question Object and the Comprehension Barrier

Referendum questions attach to specified text. Ireland is the clean example: the ballot asks ‘Do you approve of the proposal to amend the Constitution contained in the undermentioned Bill?’ and names the bill — for the 2018 amendment, the Thirty-sixth Amendment of the Constitution Bill 2018. The voter can read the exact words that will become law. The Referendum Commission, established under the Referendum Act 1998, publishes a neutral statement explaining the proposal, so comprehension pre-testing can target a fixed text. The drafting and testing steps are the subject of ballot question wording and pre-testing.

Plebiscite questions attach to a proposition: a person, a programme, a direction. Because no draft law exists, the voter cannot know what changes in law if yes wins. That is a structural comprehension barrier, not a drafting failure. A question of the form ‘do you support the government’s plan to restore order’ cannot specify what happens next, because the plan is not a law.

Mature systems pre-test. The UK Electoral Commission’s 2015 assessment found that the proposed question ‘Should the United Kingdom remain a member of the European Union?’ was unclear to a measurable share of voters, and recommended a two-option ballot — ‘Remain a member of the European Union’ / ‘Leave the European Union’ — which Parliament adopted. Canada’s Clarity Act (S.C. 2000, c. 26.8), enacted after the Supreme Court’s 1998 secession reference, goes further for secession votes: section 1 requires the House of Commons to state whether the question is clear before the vote; section 2 requires it to decide afterwards whether a clear majority exists. The Act fixes no number, which is a deliberate design choice worth its own analysis.

Parameter 3: Threshold Arithmetic

Referendum thresholds are legislated in advance and come in four families:

• Simple majority of votes cast. The default for plebiscites and many referendums.
• Turnout quorum. Italy’s abrogative referendums require a majority of the electorate to vote (Article 75). The June 2016 referendums died at roughly 30 per cent turnout; the September 2020 referendum on cutting the size of Parliament cleared 51.12 per cent turnout and passed with about 70 per cent yes.
• Approval quorum. Denmark’s constitutional referendum (Article 88) requires a majority of votes cast plus yes votes equal to at least 40 per cent of the electorate.
• Double majority. Australia’s s. 128 requires a national majority plus majorities in a majority of states; the 1999 republic referendum lost 54.9 to 45.1 nationally and failed in all six states. The Australian Capital Territory voted yes, and the ACT is not a state, so its result did not count toward the second leg. Switzerland requires a majority of voters and of cantons for constitutional texts (Article 142).

The working definitions behind these families, with the statutes that set them, are set out in quorum and threshold rules. Plebiscites typically legislate only a majority of votes cast, and the omission is where failures show up. Colombia’s plebiscite of 2 October 2016: the No side won 50.2 to 49.8 on a 37.4 per cent turnout. No turnout quorum had been set, no supermajority, no territorial weighting, and nothing in the authorising statute said what a No required. A 0.4-point margin on a 37 per cent turnout was returned into a vacuum, and the political crisis that followed was, in design terms, the threshold arithmetic arriving at an unspecified destination. That is a design flaw, not a campaign misfortune.

Parameter 4: The Implementation Path

The binding-versus-consultative line is drawn by the authorising statute, in writing, before the vote. Everything else is commentary.

UK, 2016. The vote was advisory. When the government proposed to trigger withdrawal using prerogative power, the Supreme Court held in R (Miller) v Secretary of State for Exiting the European Union [2017] UKSC 5 that an Act of Parliament was required, and the European Union (Withdrawal) Act 2018 supplied it. From the vote on 23 June 2016 to exit on 31 January 2020 ran 43 months, and a substantial share of that time went into negotiating what the question had not specified: what leaving meant. That stretch is the implementation gap, measured in time. An advisory vote on an undefined outcome hands agenda-setting power to whoever drafts the text afterwards.

New Zealand, 2020. Two votes, two designs, same election day. The End of Life Choice Act 2019 had been enacted in advance with commencement tied to the referendum result; it passed with 65.2 per cent and came into force on 6 November 2021. The cannabis referendum voted on a draft bill and was non-binding by design; it failed 50.7 to 48.2 and no further action was required. The binding difference was visible on the face of the instruments before either vote was held.

Colombia, 2016. After the plebiscite rejected the accord, the government revised it and presented the new text to Congress, which approved it at the end of November 2016 without a second vote. The result was honoured in form — the accord changed — but the mechanism was discretionary. When the implementation path is discretionary, the vote is advisory whatever the label says.

Why Plebiscites Fail Differently

A referendum in full standing has three legs: a verified initiation, a pre-legislated threshold set, and a written implementation path. Plebiscites typically lack all three, or supply them after the fact. Three failure modes follow.

Legitimacy inflation. A question about approval of a direction gets reported as a mandate for unspecified measures. Because the result cannot be audited against a text, it can be read as broadly as the winner wishes.

Sequencing inversion. Vote first, draft the implementing text later. The pathological case is the Anschluss plebiscite of 10 April 1938: annexation on 13 March 1938, ratification vote a month later, 99.7 per cent reported approval. Napoleon III’s plebiscite of December 1851 — about 92 per cent approval — is the same pattern in a different century: a vote on confidence in a person, not a law.

Unfiltered question design. With no statutory question standard and no pre-testing requirement, valence wording survives to the ballot. The Anschluss ballot printed the yes circle substantially larger than the no circle. That is not a campaign misdeed; it is ballot design, and it is auditable before the fact if anyone is required to look.

A Five-Test Classification Checklist

Run any vote through these five tests. They cite statutes, not intentions.

1. Authority test. Cite the statute that authorises the vote. Does it state the vote’s legal effect? If not, treat the vote as consultative whatever it is called.
2. Object test. Is the legal text being approved or rejected published before the vote, word for word?
3. Threshold test. Are the majority rule, quorums, and territorial majorities legislated before the campaign starts?
4. Verification test. How was the vote initiated — petition with verified signatures, legislative referral, constitutional requirement? Is the count’s chain of custody specified and observable?
5. Implementation test. On a yes, which statute commences, on what date, by which authority? On a no, what happens, and by when?

Scoring: pass all five and you are holding a referendum in full standing. Pass the first four and fail the fifth and it is a consultative referendum, with legal effect political rather than statutory. Fail the object test and it is a plebiscite; the remaining tests tell you how much design work is missing.

What the Distinction Buys You

If you advocate a vote: demand the implementation statute before the campaign, not after the result. A consultative vote donates legitimacy to whoever drafts the text afterwards, and you may not like the drafting.

If you administer one: a binding referendum requires implementation regulations drafted in parallel with the campaign. If the commencement provision does not exist on polling day, you are administering a consultation.

If you report on one: read the authorising statute first and state the vote’s legal effect in the first paragraph, not the last. The most consequential fact about the 2016 UK vote was sitting in the 2015 Act all along, and it was not printed on the ballot.

Frequently Asked Questions

Is a plebiscite always non-binding?

No. Legal effect is a property of the authorising statute, not the label. The UK’s 2016 vote was called a referendum and was advisory in law; New Zealand’s euthanasia referendum was binding because the End of Life Choice Act 2019 had been enacted with commencement tied to the result. Some votes labelled plebiscite have been given statutory effect by their authorising acts. The reliable method is to read the act and check whether it states legal effect.

Was Brexit a referendum or a plebiscite?

By label, a referendum. By design, a consultative referendum: the European Union Referendum Act 2015 specified the franchise, the question, and the count, but neither binding effect nor the legal content of leaving. The Supreme Court’s Miller judgment of January 2017 turned on exactly that gap — withdrawal required an Act of Parliament, which arrived as the European Union (Withdrawal) Act 2018.

Why can a referendum win a majority of yes votes and still fail?

Threshold arithmetic. Italy’s abrogative referendums require a majority of the electorate to vote (Article 75); the June 2016 referendums died below that quorum despite majority support among those who voted. Denmark’s Article 88 requires yes votes equal to at least 40 per cent of the electorate. Australia’s s. 128 requires a national majority plus majorities in a majority of states. The threshold set is part of the question, not an afterthought.

Who decides whether a ballot question is clear?

It varies by system, and the variation is the finding. Ireland attaches every constitutional referendum to a named bill, so the question is clear by construction. The UK Electoral Commission has a statutory role in testing question intelligibility under the Political Parties, Elections and Referendums Act 2000, and its 2015 assessment changed the EU referendum ballot. Canada’s Clarity Act assigns the clarity judgment to the House of Commons — before the vote for the question (section 1), after the vote for the majority (section 2). Where no body or standard exists, question quality depends on the government that drafted it, which is the plebiscite condition.

Can a government ignore a vote it called?

Legally, if the vote is consultative, yes — the cost is political. Colombia’s government revised the rejected accord and obtained congressional approval weeks after the 2016 plebiscite. The UK Parliament legislated to implement a result the 2015 Act had not bound it to implement. If you want a vote that cannot be ignored, enact the implementing statute before the vote, as New Zealand did in 2019.

Next on this site: a working glossary of threshold terms — turnout quorum, approval quorum, double majority, supermajority — because threshold arithmetic is where most classification arguments actually live; and a piece on signature verification chains, from validity rates to the deadlines that make a petition auditable. If you have a vote you want run through the five tests, send the authorising statute and the ballot question. It takes about twenty minutes, and the output is a case file.