the numbers guy

Archive for April 2012

Well, that was unexpected! Nobody anticipated last night’s result. Even the most optimistic of PC supporters were hoping for, at the very best, a slim majority. Not a single pundit or political analyst predicted a PC landslide on the scale Alison Redford pulled off last night.

Every single pollster also got it wrong. Sunday’s final Forum Research poll did capture a last minute swing to the PCs – but still had the Wildrose ahead 38-36. On the night, the PCs won with nearly a 10 point lead over Wildrose.

With pollsters unanimous in predicting a Wildrose victory, it’s not surprising that the seat projection models were well off the mark too. The closest was ThreeHundredEight, not least because its final projection placed significant weight on Sunday’s Forum Research poll.

Still, even ThreeHundredEight projected a Wildrose victory – although you can almost sense the uncertainty in nearly every paragraph of 308 author Eric Grenier’s  Final Alberta Projection. Mr Grenier, more than anyone else I read in the hours before polls closed, seemed to sense that something was in the air and that a PC win might be on the cards.

My own seat projection model relied more heavily on Friday and Saturday’s polling figures which unanimously had Wildrose well ahead – my seat projection was therefore out by a country mile. I’ll get on to some of the reasons why – and the lessons learned – in a moment, but first here are some obligatory humble pie charts:

Political analysts will speculate for months about the reasons behind the last minute withering of the Wildrose surge – Josh Wingrove makes a start in this morning’s Globe and Mail.

From my perspective, I’m more interested in learning more about how the pollsters could have been so wrong. There was a 9.5% lead for the PCs on the night. The most favourable final opinion poll for the PCs had them 2 points behind. Such a huge differential in a heavily-polled major election contest is almost unprecedented in recent times.

The final numbers from most other pollsters had the PCs even further behind. A Wildrose majority seemed all-but-certain, a landslide highly possible. I can’t recall a provincial or federal election – in Canada or even in other Commonwealth countries – in which the pollsters were so far off the mark in terms of the final result. Even with last year’s federal election, while few predicted a Tory majority, some of the final polls did pick up the last minute Tory surge in Ontario’s 905 region.

My focus will also be on improving my seat projection model as we look ahead to the next provincial elections in British Columbia and Quebec. The model was basically designed to translate polling figures into seats. Polling figures that are way off the mark will inevitably translate into seat projection figures that are also way off the mark.

Nevertheless, next time I will pay much closer attention to the last minute final-day polls and will weight them accordingly. Among other initial ‘lessons learned’ are the need to incorporate an adjustment for strategic voting – where supporters of minor parties end up shifting their vote to a leading party in the final hours (or indeed on election day itself).

A further factor was how well incumbents performed last night – particularly cabinet ministers and also well-entrenched minor party incumbents such as the Liberals’ David Swann in Calgary-Mountain View and Laurie Blakeman in Edmonton Centre. Swann and Blakeman each bucked the province-wide trend for their party, winning their ridings comfortably despite predictions they were in danger of losing their seats. Adjustments will need to be made to future seat projection models to reflect this.

All in all, it was an absolutely fascinating election night – and analysts, pollsters, pundits and political junkies from coast to coast to coast will be poring over the evidence from Alberta 2012 for many months ahead.

Advertisements

Alberta’s hard-fought provincial election takes place tomorrow (Monday 23 April). It has shaped up as a head-to-head contest between the governing Progressive Conservatives (who have been in power for over 40 years) and the right wing conservative Wildrose Party. Wildrose surged into a strong early lead in the campaign, but some polls taken over recent days have hinted that the race may be tightening in the final stretch.

Alberta is the heartland of Canadian conservatism and, while the NDP and Liberals will be a factor in Edmonton and, to a lesser extent Calgary, neither is in the hunt for seats in rural Alberta. The small, centrist Alberta Party and the EverGreen Party are also contesting the election.

For last fall’s provincial elections in Prince Edward Island and Saskatchewan, this blog confined itself to setting the scene – looking at where the close contests might be and so on. For the Alberta election, I’m going to go a step further, and attempt to project the results, riding by riding, based on the most recent polling evidence available.

Constructing a reasonably accurate seat projection model for Alberta 2012 presented a number of challenges. A recent boundary reviewincreased the number of seats in the legislature from 83 to 87. There was a great deal of shuffling around of electoral boundaries, particularly in urban areas, which made a straight projection of districts from the 2008 results difficult. Where boundary changes were significant, I made a best estimation based on previous poll-by-poll results, as to how the 2008 election would have gone if contested on the new 87-district boundaries.

I also made a few further tweaks, reflecting the likely impact of high profile defections, such as that of Guy Boutilier from PC to Wildrose in Fort McMurray-Wood Buffalo. For seats where Wildrose didn’t stand a candidate in 2008, I assigned them a notional support value, reflecting that of their weakest 2008 electoral districts in that region. I didn’t assign an incumbency bonus for sitting MLAs – but I did give a bonus for the party leaders.

All this resulted in a set of baseline figures for each of the 87 electoral districts and, after some deliberation, I opted for a proportional swing model to project each result. I had some reluctance in doing so, as a weakness of this kind of model can be in an election which sees a party surge from almost nowhere at the previous election. This was the case with the NDP in Quebec in last year’s federal election, and it looks like being the case with Wildrose in this election – who are leading in the polls at around 40% despite polling just 6.8% at the previous election (and who didn’t even contest over a quarter of the seats in 2008).

Nevertheless, a proportional model is far superior to alternatives such as uniform swing – so we’ll see how it fares when the results are in. We’re likely to face a similar ‘surge from almost nowhere’ situation in British Columbia next year with the rise of the BC Conservatives and, if necessary, I’ll make adjustments to my BC model to reflect this.

The best seat prediction model in the world will fail if the input information (the polling evidence) is inaccurate. The numbers I fed into the model came from aggregating the results of the final surveys of pollsters who have been regularly polling the race. These numbers were broken down into three regions – Calgary, Edmonton and the Rest of Alberta – with a view to capturing localized trends, such as the apparent steep decline in Liberal support in the Calgary region.

The results are at the top of this thread – a solid majority for Wildrose who are projected to win 55 seats across the province. The Progressive Conservatives are projected to be reduced to just 27 seats – a huge drop from the 72 they won at the 2008 election. The Liberals are narrowly projected to pick up a single seat in a very tight four-way race in Edmonton Centre. The NDP are projected to win 4 seats – by picking up two extra seats in Edmonton.

  • Regionally the model projects Calgary to return 22 Wildrose and 5 PCs.
  • In Edmonton, it projects 13 PCs, 11 Wildrose, 4 NDP and 1 Liberal.
  • In the rest of Alberta, Wildrose is projected to win 22 seats, with the PCs winning 9.

I’m looking forward to seeing how these figures – and my headline total of Wildrose 55, PC 27, NDP 4, Lib 1 – compares with reality when the results come through on Monday night.



  • None

Twitter – numbersguyblog

Error: Twitter did not respond. Please wait a few minutes and refresh this page.