Research · When the polls missed

The Wildrose lead that vanished overnight: Alberta's 2012 polling miss

Abstract. The final polls averaged Wildrose 41, Progressive Conservatives 34; the PCs won 44.0% to 34.3% and kept a 61-seat majority. Pollsters blamed a late swing of undecided and centre-left voters to the PCs after most firms had stopped polling, made worse by a 14-point jump in turnout.

Polling Forecast · 2 November 2026 · 6 min read · Canada, 2012 Alberta provincial election, polling error, final polls, inquiry, pollster methods

Summary

Every published poll of the 2012 Alberta campaign had Wildrose ahead. The final poll from each of the seven firms averaged Wildrose 41.0, Progressive Conservatives 34.0: a 7-point lead. On 23 April the PCs won 44.0% to 34.3% and took 61 of 87 seats. The error on the lead was 16.7 points. The pollsters' own explanation was a late swing: voters, many of them Liberals and New Democrats, moved to Alison Redford's PCs in the last two or three days to stop Danielle Smith, after most firms had stopped polling.

What the polls said

The last poll from each firm with fieldwork ending within ten days of polling day (figures as tabulated from the firms' releases, most of which are no longer online):

Pollster	Fieldwork end	PC	Wildrose	Liberal	NDP
Forum Research	22 Apr	36	38	10	12
Angus Reid	21 Apr	32	41	13	11
Campaign Research	19 Apr	34	41	11	11
Abacus Data	19 Apr	31	41	12	13
ThinkHQ	18 Apr	33	41	11	11
Léger	16 Apr	36	42	9	10
Return on Insight	14 Apr	36	43	11	9
Average		34.0	41.0	11.0	11.0

Wildrose had led by more earlier. Abacus had it 17 points ahead (46 to 29) in the poll that ended on 11 April, the night before the leaders' debate. The race tightened in the last week, but no poll in the campaign put the PCs in front.

What happened

Elections Alberta's official figures: PC 43.97%, Wildrose 34.28%, Liberal 9.89%, NDP 9.85%, Alberta Party 1.31%. Seats: PC 61, Wildrose 17, Liberal 5, NDP 4. Turnout rose to 54.37% from 40.59% in 2008, the first rise since 1993.

Against the average, the PCs were 10.0 points too low and Wildrose 6.7 too high. The Liberals (1.1 too high) and the NDP (1.2 too high) were close. Even Forum, the only firm in the field over the final weekend, missed the lead by 11.7 points.

Why it happened: what the pollsters found

No industry body held an inquiry. The Marketing Research and Intelligence Association (MRIA) ran no review of these polls. The explanations came from the firms themselves, mostly in the days after the vote.

Angus Reid (Mario Canseco, 24 April 2012). The polls would have been right if the election had been held on the Saturday. The firm closed its fieldwork on Saturday morning to publish in time and "saw that shift drastically in the final day and half of the campaign". He said he had never seen all the pollsters get it this wrong.
Léger (Ian Large, 26 April 2012). Léger ran an online survey of 481 voters the day after the election. 39% said they had decided in the final week or later: 11% over the last weekend, 7% on election day and 7% in the polling booth. Among PC voters it was 45%, among Wildrose voters 26%. 27% of PC voters said their main reason was to block another party. Large said that "could almost entirely explain" the gap with the final polls. 35% of PC voters said they had considered Wildrose.
ThinkHQ (Marc Henry) noted how few outliers there were and suggested overnight tracking with a rolling average. Janet Brown, whose model had projected a Wildrose majority, said "public opinion changed dramatically, quickly and right at the end of the campaign".
Return on Insight (Bruce Cameron) blamed online and automated phone polls, which he said were weaker than live interviewers. Allan Gregg of Harris-Decima pointed to the strength of incumbency, which he said pollsters underrated.

The PCs' campaign said its internal polling had it ahead throughout. Wildrose's late trouble came from two candidates' remarks: Allan Hunsperger's "lake of fire" blog post on gay people and Ron Leech's comment on his advantage as a white candidate. Both surfaced in the final week of the campaign.

The fullest later study, by David Coletto (Abacus) and Bryan Breguet in the Canadian Political Science Review (2015), found late deciders alone could not account for it. Even the top of the PC range in Abacus's volatility data fell short of the 44% the party won. Across nine provincial elections, the change in turnout was the strongest predictor of poll error, and Alberta's turnout jumped nearly 14 points.

What our data shows

Our reading of the published figures:

The miss was in the governing party's share. The PCs were understated by 10 points, more than Wildrose was overstated (6.7). The Liberal and NDP shares were nearly right, so the late movement came from both sides: soft Wildrose voters went back to the PCs, and some centre-left voters lent the PCs their vote. That fits Léger's 27% blocking vote.
Turnout changed the electorate. Valid votes rose from 950,363 to 1,290,352. Most of the extra votes went to Wildrose (up 377,918), but the PCs' vote also rose by 66,249 even as their share fell 8.8 points, while the Liberal vote fell by 123,532. A turnout rise this large is exactly what a likely-voter screen built on past turnout struggles with.
Timing explains some, not all. Forum's two final polls moved from Wildrose +9 (21 April) to Wildrose +2 (22 April), so there was real movement at the end. Forum was still 11.7 points out on polling day.
Every firm missed the same way. Online, IVR and phone polls all had Wildrose ahead. Mode made no difference to the direction.
Against the federal record it is enormous. At Canadian federal elections since 2000, the largest error on the gap between the top two parties is 5.8 points (2004) (Canada polling error paper). Alberta 2012 was nearly three times that. A year later British Columbia produced an 11.5-point miss, with the incumbent again understated.

What the pollsters changed

There was no industry review and no common method change. The changes were firm by firm and mostly about timing:

Angus Reid said it would poll "until 11:59" before future elections, not stop days early.
ThinkHQ proposed rolling overnight tracking to catch fast shifts. Léger's Large expected more use of online polling, which he called "quick" and "just as accurate".

The changes did not stop a repeat. Thirteen months later the British Columbia polls missed the same way. In their review of the 2015 Alberta election, Janet Brown and John Santos rated the polls only "marginally better" than in 2012, which they said failed all four of their accuracy tests.

What it means for our forecast

Our Canadian correction rests on federal elections only. Half of each party's average past miss is taken off today's polls, and the rule is kept because it beat the plain average at seven of nine federal elections when each was predicted without knowing its result (Canada polling error paper). Provincial misses like Alberta 2012 are not in that record, and a correction learned from one province's single election would not pass the same test.

For the provinces we forecast, the simulations instead draw a province-wide error for each of the two main parties with a standard deviation of 4 points by default, on top of regional and district noise. A 10-point PC miss is two and a half standard deviations, so it sits in the tail of the simulations, not outside them. Firm-level records are on our Canadian pollster ratings.

Sources

How to cite

Polling Forecast (2026). The Wildrose lead that vanished overnight: Alberta's 2012 polling miss. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-ca-2012-ab. Live figures change as new polls arrive; the date above is when the text was last revised.