Referendums in the UK

A referendum puts a single question directly to the electorate. Voters decide the answer themselves instead of delegating it to the MPs they elected. It is the main element of direct democracy in an otherwise representative British system.

The UK has no codified constitution and no rule requiring a referendum on anything. Each one therefore needs its own Act of Parliament, which sets the question, the franchise, the date and the legal effect. That gives the government of the day very substantial control:

• The government chooses whether to ask at all. No mechanism exists for voters to force a national referendum on Parliament. Switzerland and several US states have an initiative for exactly that; the UK does not.
• The government shapes the question, subject to review by the Electoral Commission, created by the Political Parties, Elections and Referendums Act 2000. The Commission assesses whether a question is intelligible and unbiased, designates the lead campaign on each side, and enforces spending limits.
• The government chooses the legal effect. Most UK referendums are advisory: parliamentary sovereignty means the result cannot bind Parliament in law. A minority are binding, because the enabling Act says the outcome must be implemented.

The advisory/binding distinction is worth learning precisely, because it is the pivot of almost every exam question about referendums and sovereignty. Advisory does not mean weak. It means legally non-compulsory and politically near-irresistible.

Before 1997, referendums were rare enough to be treated as constitutional oddities.

(Extra context — the specification asks about referendums since 1997. The two earlier votes below are background that explains why 1997 changed the habit, not assessable content in themselves.)

The 1975 referendum on continued membership of the European Communities was the first held UK-wide, and Harold Wilson called it largely to manage a split in his own party. In 1979 devolution referendums in Scotland and Wales were wrecked by a statutory rule requiring 40% of the whole registered electorate to vote Yes: Scotland produced a narrow Yes majority that fell short of the threshold, and Wales rejected devolution outright.

The Blair government elected in May 1997 arrived with a large constitutional reform programme and used referendums as its delivery mechanism. Within eighteen months it had put devolution to voters in Scotland, Wales, Northern Ireland and London. From that point a convention developed: significant constitutional change is put to a public vote first.

Notice the pattern: a dense cluster of devolution votes at the start, a long gap, then three high-stakes national questions in five years.

The four devolution referendums are the clearest illustration of what referendums have been used for in the UK: legitimising constitutional change that a government has already decided to make.

Scotland, 11 September 1997. Two questions on one ballot. 74.3% backed a Scottish Parliament and 63.5% backed giving it tax-varying powers, on 60.2% turnout. The margins were wide enough that the Scotland Act 1998 passed with the devolution settlement treated as settled.

Wales, 18 September 1997 — the case study worth knowing in detail. Wales voted Yes by 559,419 to 552,698, a majority of 6,721 votes, or 50.3% to 49.7%, on a turnout of just 50.1%. On those figures barely a quarter of the Welsh electorate actively endorsed devolution. Two lessons follow:

• A referendum can legitimise a change on a very thin base. The Senedd's existence has not been seriously reopened since, which suggests the legitimacy conferred was durable regardless of the margin.
• It is an argument for thresholds. Had the 1979 rule requiring 40% of the whole electorate still applied, Wales would have voted Yes and got nothing.

Northern Ireland, 22 May 1998. 71.1% endorsed the Good Friday Agreement on 81.0% turnout, the highest at any referendum held in Northern Ireland. A parallel referendum in the Republic of Ireland passed on the same day. Here the referendum did work no election could: it gave a cross-community peace settlement the direct consent of the people bound by it.

London, 7 May 1998. 72.0% backed a directly elected mayor and assembly, but on 34.1% turnout — a large Yes on a small vote, and a standing reminder that a lopsided result is not the same thing as an engaged electorate.

And a No. In November 2004 the North East of England rejected a regional assembly by roughly 78% to 22% on 47.7% turnout. The result stopped regional devolution in England, and no further regional assembly referendum followed. Referendums can close a question as firmly as they open one.

The AV referendum of 5 May 2011 is the exception on legal effect, and the best worked example of a binding UK referendum.

Origin. It was the price of coalition. The Liberal Democrats entered government in May 2010 committed to proportional representation; the Conservatives were opposed. The compromise written into the coalition agreement was a referendum on the Alternative Vote — a preferential system that neither party had campaigned for — held under the Parliamentary Voting System and Constituencies Act 2011.

Legal effect. That Act required ministers to bring the change into force if voters said Yes. There was no further parliamentary decision to be taken, which makes it the clearest example of a UK referendum that bound the government by statute.

Result. 67.9% No to 32.1% Yes, on 42.2% turnout. Every nation of the UK returned a No majority.

Impact, which is what the specification asks for:

Consequence	Detail
Electoral reform removed from the agenda	Reformers had spent a century seeking a national vote; losing it two-to-one closed the argument for a generation
Coalition relations soured	The campaign was fought bitterly between two governing parties, damaging trust inside the coalition
A template for binding referendums	Parliament showed it can pre-commit to a result by statute, which is why the advisory/binding distinction is a real choice rather than a formality
An argument against referendums	Turnout of 42.2% meant roughly 29% of the electorate delivered the No verdict that settled the question

A common trap: AV was not on the 2026 specification's list of systems and it is not used anywhere in the UK. Know it as the subject of a referendum, not as a system you must evaluate.

Held on 18 September 2014 under the Edinburgh Agreement between the UK and Scottish governments, this asked "Should Scotland be an independent country?"

Result. No 55.3%, Yes 44.7%, on a turnout of 84.6% — the highest recorded at any UK-wide or nation-wide vote in the modern era. The franchise was extended to 16- and 17-year-olds for the first time in a major UK poll, a precedent Scotland and Wales later made permanent for their own elections.

Impact on UK political life:

• It did not settle the question. Both governments described the vote as a once-in-a-generation event, yet independence remained the central axis of Scottish politics afterwards.
• It transformed party competition. At the 2015 general election, months after losing, the SNP won 56 of Scotland's 59 Westminster seats.
• It produced further devolution. The pre-referendum promise of additional powers, made jointly by the three main UK party leaders, led to the Smith Commission and then the Scotland Act 2016.
• It set a legal boundary. In November 2022 the Supreme Court held, on a reference by the Lord Advocate, that the Scottish Parliament has no power to legislate for an independence referendum without Westminster's agreement. A second referendum therefore requires the consent of a UK government that has no incentive to grant it.

The 2014 vote is the strongest single piece of evidence for referendums as participation-boosters, and the strongest single piece of evidence that they do not close questions permanently.

The referendum of 23 June 2016 produced 51.9% Leave to 48.1% Remain on 72.2% turnout, with 17.4 million Leave votes — the largest number ever cast for anything in the UK.

The legal position was the opposite of what most voters assumed. The European Union Referendum Act 2015 contained no obligation to act on the outcome, so the vote was advisory. That mattered almost immediately.

Worked case study — R (Miller) v Secretary of State for Exiting the European Union [2017] UKSC 5. The government argued it could trigger Article 50 using the royal prerogative, on the authority of the referendum. The Supreme Court held by 8 to 3 that withdrawal would strip away rights created by an Act of Parliament, and only Parliament can remove what Parliament has granted. Ministers had to obtain the European Union (Notification of Withdrawal) Act 2017 first.

The ruling drew the line exactly: a referendum changes the politics, and an Act of Parliament changes the law.

Impact on UK political life:

• Leadership. David Cameron resigned the morning after the result.
• Party realignment. The referendum cut across both main parties, split the Conservatives for years, and opened the space on the right that Reform UK occupies. The Financial Times noted in May 2026 that the realignment of politics on the right since Brexit was still being cemented.
• Parliament against the people. The 2017–19 Parliament was dominated by the argument over whether MPs elected as representatives could revise a decision voters had taken directly.
• A constitutional precedent nobody planned. A 51.9% vote on a single day produced the largest change to the UK's legal order since 1973, with no supermajority requirement and no threshold.

As of September 2026 it remains the most recent UK-wide referendum.

Case for	Case against
Consent from the people a change binds. Devolution moved power over whole populations, and asking those populations directly is a different act from asking their MPs.	No threshold and no supermajority. A single day's simple majority carries the change: Wales approved devolution in 1997 on the active support of about a quarter of its electorate.
Settles issues that cut across parties. Devolution, EU membership and electoral reform sit awkwardly in a party system built on economic divides.	Blunt instrument. A yes/no question cannot capture the range of options; "Leave" covered several incompatible destinations.
Every ballot weighs the same. Under FPTP a vote cast in a safe seat changes nothing. A referendum is counted as one national or nation-wide total, so where a voter lives does not decide whether their vote counts.	Turnout can be low. 42.2% in 2011 and 34.1% in London in 1998 mean small minorities of the electorate decided lasting questions.
Entrenches constitutional change. Devolution is politically irreversible because voters, not just MPs, approved it.	The mandate stops at the question asked. Voters chose to leave the EU, not the terms of leaving, so every consequential decision returned to the Parliament the vote was meant to bypass.
Checks executive power. A Commons majority can pass almost anything, but a referendum puts the decision beyond the government's control, as David Cameron found in 2016.	Government controls the ask. Ministers choose whether, when and what to ask; referendums are as often a tool of party management as of popular control.
Popular and expected. The convention is now strong enough that omitting a referendum on a major change would itself be controversial.	Legal effect is fixed by the enabling Act, not by the vote. The ballot paper does not say whether the result binds anyone; in 2016 that had to be answered after the event, by the Supreme Court in Miller.

Local practice complicates both columns. Since the Localism Act 2011, an English council proposing a council tax rise above a set threshold must put it to a local referendum, which has made referendums a routine constraint on local government rather than a constitutional set-piece. When ten English cities voted in 2012 on whether to adopt directly elected mayors, nine said No — only Bristol said Yes. Referendums are used most often where the stakes are lowest.

A strong judgement rarely says referendums are simply good or bad. It says what they are good for: conferring legitimacy on a change a government has already chosen, and settling a question that divides parties internally. It then asks whether the price — blunt questions, unprotected minorities, and results decided by turnout as much as opinion — is worth paying for that.

1. Writing only about 2016

Section 3.2 asks how referendums have been used since 1997, which covers devolution, the AV referendum, Scottish independence and local referendums. An answer built entirely on Brexit has covered one ninth of the evidence.

2. Calling every UK referendum advisory

Most are, but the 2011 AV referendum was binding under the Parliamentary Voting System and Constituencies Act 2011. Stating the general rule and naming the exception shows you understand why the distinction exists.

3. Confusing a referendum with an initiative

A referendum is called from above by government and Parliament. An initiative is triggered from below by citizens collecting signatures, as in Switzerland and several US states. The UK has the first and not the second, and that asymmetry is itself an argument in the debate.

4. Treating high turnout as a general feature

2014 reached 84.6%; 2011 managed 42.2% and the 1998 London vote 34.1%. Whether referendums raise participation depends entirely on the question, so use the range rather than one flattering figure.

5. Assuming referendums settle questions

Scotland voted No in 2014 and independence stayed at the centre of Scottish politics; the 2022 Supreme Court reference exists because the argument continued. Judge each referendum on whether it closed the issue, and explain why some do and some do not.

6. Leaving the judgement to the last line

The 30-mark command word is Evaluate, and a third of the marks are for AO3. Signal your line in the opening paragraph, return to it after each section, and finish with a conclusion that follows from what you have argued rather than announcing a new position.