Result · Thursday, 18 June 2026

Aberdeen South by-election

Conservative gain from SNP. Douglas Lumsden (Conservative) won on 49.5%, a 6,050-vote majority (+20.9pp) over Richard Thomson (Scottish National Party, 28.6%). Turnout 37.4%. Scotland · S14000061.

Elected MP Douglas Lumsden Conservative

Declared result (vote share)

  • Conservative 49.5% Douglas Lumsden
  • Scottish National Party 28.6% Richard Thomson
  • Reform UK 8.6% Jo Hart
  • Labour 5.4% Nurul Hoque Ali
  • Lib Dem 4.4% Mel Sullivan
  • Green Party 3.4% Jorg Shelton-Eckstein
  • Other 0.2%

Trigger

Why this seat was contested

Stephen Flynn (Scottish National Party) Flynn was elected to the Scottish Parliament on 7 May 2026; the Scottish Elections (Representation and Reform) Act 2025 bars a dual Holyrood+Westminster mandate, so he resigned the Commons seat on 14 May 2026.

Declared result · 18 June 2026

Full result

PartyVotesShareSwing vs GE2024
Conservative 14,308 49.5% +25.1pp
Scottish National Party 8,258 28.6% -4.2pp
Reform UK 2,478 8.6% +1.7pp
Labour 1,550 5.4% -19.3pp
Lib Dem 1,270 4.4% -1.9pp
Green Party 974 3.4% -0.1pp
Other 59 0.2% -1.2pp
Majority6,05020.9ppn/a
Turnout28,89737.4%n/a

Source: returning officer declaration via primary record.

The signal · no constituency polls existed

What the freshest data pointed to

No published constituency polling for this by-election (confirmed by PollCheck and Ballot Box Scotland). By-elections get no broadcaster exit poll either. The signal below is the substitute.

Pre-election call: Ballot Box Scotland: Lean SNP.

7 May 2026 Holyrood overlap

Aberdeen Deeside and North Kincardine (Holyrood constituency). Stephen Flynn won this Holyrood seat on 7 May 2026, the win that triggered the Westminster by-election. Different electorate to the Westminster seat; constituency (first) vote.

  • Scottish National Party 34.1%
  • Conservative 30.5%
  • Reform UK 17.7%
  • Lib Dem 8.3%
  • Labour 8.1%

GE2024 baseline (4 Jul 2024)

The year-old general-election result, a naive prior.

  • Scottish National Party 32.8%
  • Labour 24.7%
  • Conservative 24.4%
  • Reform UK 6.9%
  • Lib Dem 6.3%
  • Green Party 3.5%
  • Other 1.4%

Scorecard · how every forecaster did

Forecasts vs result

ForecasterTypePredicted winnerShare MAECall
Actual result declared Conservative n/a n/a
UK Elections Holyrood signal · 7 May (basis) signal Scottish National Party 8.1pp
GE2024 result (naive prior) · 4 Jul 2024 baseline Scottish National Party 8.7pp
Electoral Calculus (GE seat model) · 9 May seat model Scottish National Party 9.0pp
UK Elections GE seat model · 19 May seat model Scottish National Party 9.6pp
Betting markets (eve of poll) · 17 Jun market Scottish National Party (79%) n/a
Ballot Box Scotland (qualitative) · 4 Jun qualitative Scottish National Party n/a

How every forecaster did, graded against the declared result. No constituency poll or bespoke by-election model existed for this seat; the entries are the freshest signal, the Electoral Calculus GE-model seat estimate, betting markets, and the Ballot Box Scotland qualitative call. MAE is over each forecaster's reported parties.

An upset. Every area signal pointed to an SNP hold: the GE2024 baseline (SNP 32.8%), the 7 May 2026 Holyrood overlap (SNP 34.1% ahead of the Conservatives' 30.5%), and Ballot Box Scotland's 'Lean SNP' call. The result was a 21-point Conservative GAIN, the first Conservative gain of a Westminster seat at a Scottish by-election since 1967.

The mechanism was tactical unionist consolidation on a low (37%) by-election turnout. Labour collapsed from 24.7% to 5.4% (-19pp) and the Lib Dems eased, with that vote coalescing behind Douglas Lumsden to beat the SNP. The SNP's own share barely moved (-4pp); they were not so much defeated as outflanked by a unified anti-SNP vote.

Reform UK is the signal's other big miss: 17.7% in the Holyrood overlap five weeks earlier, just 8.6% here. Reform's Holyrood list strength did not transfer to a Westminster by-election framed as a straight SNP-vs-Conservative fight. No area-level signal could have called this; it needed a model of tactical switching that the raw shares do not contain.