Yes, there is a science to predicting elections

By Lethbridge Herald on May 2, 2025.

Allan Wilson
For the Herald

Sunday night I predicted almost exactly what happened with our federal election:  Liberal minority that is very close to a majority.  And the local Liberal would get more votes than any opposing candidate in decades but still lose. There are five major polling companies in Canada. Not one predicted a Conservative victory; three predicted a Liberal majority, and they came within three seats of that result.

Since 2015 I have made predictions for four Canadian federal elections and the Lethbridge riding, the 2015, 2019, 2023 Alberta elections and seven Lethbridge ridings (includes the 2024 Lethbridge by election). And the 2016, 2020 and 2024 US elections, and the 2018 and 2022 US mid terms.

In those 23 elections and I have picked the winner 21 times. I missed the two times Trump won.

It wasn’t luck or great brainpower. It was sorting the results of reputable polls, and researching observations from experienced writers. And in the case of the seven Lethbridge ridings in Alberta elections, it is my own system of analyzing lawn signs.

Reputable polling companies want to be right and try to balance possible bias with research into scanning the social landscape for the proper mix of demographics about gender, location, age and voting history. If they make good information their readers can make their own predictions.  They want to be right so that when the election is over they can continue to predict how consumers would like a product or service or policies…in other words, stay in business.

Good polling companies will show in detail their methodology. That is how I discovered a polling site in U.S. elections that always seemed to have Trump three points ahead of other polls.  It turned out the site was created by a former staff member for George Bush, and they were only using land lines. And there was a poll paid for by an Iowa newspaper that said Harris was ahead by seven points in their state, so different from other polls. And in the last days of an election in both Canada and the US new polling companies appear, but they always turn out to be fake creations designed for cheerleading purposes. Last Friday there were two different polling sites appeared on You Tube that insisted the Conservatives were tied with the Liberals, significantly different from the other five established polls.

Are polls useful?  Yes. They help to raise money for the candidates, and they bring in volunteers. They can steer campaigns with their results.  They can build or stall momentum with voters. They create talking points with the media that can build voter turnout.

In order of reliability, what are the most accurate pole methods?

1. Human questioning a voter on the phone.

2. Automated voice questioning a voter on the phone.

3. Online polling from a demographically accurate pool (they contact you).

4. Online polling where anyone can give their opinion (it can be easily be manipulated).

5. Reputable polls always cloak/protect their numbers with the +2 or -2 point margin, 19 times out of 20. With the 23 elections I made predictions for, the respected polls were always within 2.5 points of the final result. With Trump beating Clinton the polls had Clinton ahead by only 1.5 points, and with Biden beating Trump, it was +2.5 points.  With Trump beating Harris in 2024, it was a 1-2 point Harris lead. In both Trump victories those predictions were within the 2 point margins:  Trump won three swing states of Pennsylvania, Michigan and Wisconsin by 88,000 votes in 2016, and in 2024 won those same swing states by 144,000 votes, very small numbers compared to the hundreds of millions of votes.

 As for the Lethbridge provincial elections, in 2015 the NDP lawn signs were appearing in greater numbers than I ever saw before. In Lethbridge East I called Maria Fitzpatrick, the NDP candidate, and she confirmed that she was getting great results knocking on doors, and I heard similar stories in Lethbridge West, and it looked like a historic turnaround.  And it was.  

I continued with the 2019 and 2023 elections and the 2024 byelection.  I made the correct predictions in all seven elections.  This is what I learned:

1. Signs on public boulevards don’t count because anyone with money can put them up.

2. Lawn signs don’t tell you what voters in apartments and senior living communities are thinking.

3. Lawn signs almost always represent two voters (spouses).

4. Lawn signs show the committed voters who will actually vote.

5. Lawn signs are excellent advertising.

6. Candidates don’t knock on doors with signs on the lawn – the voters have made up their minds.

7. At the end of the campaign, if the candidate has 60 per cent of the lawn signs…they win.  They don’t necessarily get 60 per cent of the vote because of the limiting factors in point No. 2, but they win.  The lawn signs indicate the strength of the candidate ground game.

 With neutral reasoning and the developing science of polling with balanced demographics, predicting elections can be accurate (given or take 2 points!).  But after the election….if only governing the province and the country would be that easy.

Allan Wilson is a retired teacher involved in Lethbridge politics and music since 1990.   He has been published in the Toronto Star, Reader’s Digest, Globe & Mail and the Canadian Medical Association Journal.