Polls Can't Predict Conservative Support

October 3rd, 2021 | DS

Opinion polls have been notorious in their inability to accurately predict Conservative Party support. Some polling agencies have remained in business through the past three elections by botching their predictions to the highest degrees, while only a small handful have come closer to accuracy. This is a rundown and analysis of the polls from the past two elections.


Conservatives Finished 2-3% Higher Than Predicted

The polling average for the Conservative Party was 31% for a majority of the 2021 campaign. When the final numbers were counted after election day, the party finished near 34%. The polling averages and the election day finals for Conservatives in 2019 were identical. The CPC polled at an average of 31% but finished the election at 34%.

This has been part of an ongoing trend reaching further back than the past two federal elections. In 2011, Conservative support was pegged at an average of 35%, but the party finished at 39% and won a majority. Prior to election day, media was using the polling data to predict another Conservative minority. In the days leading up to the election, EKOS pegged Conservative support at 34%, one of the lowest of all polling agencies. In 2021, EKOS pegged Conservative support at 28% on the eve of the election.

On the eve of the 2008 federal election, EKOS put Conservative support at 35%, but the party finished at 37.6%.

One of the most accurate polling agencies of the past two federal elections has been Leger. In 2019, they put Conservative support at 33% just days before the election, missing the mark by a single digit. In 2021, they did the same in predicting Conservative support at 33% and Liberal support at 32%. This was one of the most accurate predictions, to date, by any national polling agency.


Who Owns The Pollsters?

EKOS is one the most persistent polling agencies in Canada and has attempted to measure results for the past seven elections. In the 2000 election, EKOS Research predicted a 45% win for Jean Chretien's Liberals, but the party finished at 40%. Ever since, with the exception of 2015, EKOS has measured Conservative Party support at much lower levels than any other national polling agency.

Frank Graves is the president of EKOS Research Associates and has been accused of being a prominent Liberal Party donor and supporter. After the 2021 results, journalist Brian Lilley stated, “I”ve never seen a pollster get it so wrong and then spin so badly.”


(article continues after ads)

Following a tsunami of criticisms and attacks, Frank Graves made his Twitter profile private. Leading up to the election, Graves boldly and brashly predicted a Liberal win in Ontario, while casting doubt on the Conservative Party's ability to win any seats. Days prior to the election, Graves tweeted, “The LPC lead in Ontario is so large that I'm not even going to report it. I am not sure there are any safe CPC seats in Ontario.”

Mainstreet Research, which pegged Conservative support at 30% on the day before the 2021 election, is run by Quito Maggi. Not much is known about his political affiliations and Mainstreet's track record leaves no indication of bias. The company has successfully pegged Conservative wins in the past and Maggi has no record of exhibiting a lean toward any side of the political fence. However, Mainstreet failed to accurately predict Conservative support in 2021. 

Abacus Data, which pegged Conservative support at 32% and Liberal support at 31%, is a fairly new polling agency founded in 2010. The company was founded by David Coletto, who runs the daily operations at Abacus. He has a doctorate from the University Of Calgary and has no apparent and visible political biases. Coletto's firm was one of the more accurate pollsters of the 2021 election.

Ipsos is a multi-national company that is traded publicly. Their predictions were some of the most accurate in the past four elections. They are often commissioned by media and large corporations for market and political research.

Angus Reid is owned by Alida, which is a prominent software company based in Toronto. The CEO of Alida is Ross Wainwright, who has no publicly available political affiliations. On social media, critics have called Angus Reid a “conservative think-tank”, which is unsubstantiated and inaccurate. These criticisms stem largely from the 2021 English leaders debate, which was hosted by the Angus Reid Institute. The criticisms are linked to Liberal partisans who found themselves frustrated with the debate's moderator, Shachi Kurl.

Lorne Bozinoff is the founder of Forum Research, which has often released polls that slightly favour Conservatives but that have accurately predicted many elections, including municipal races. Forum Research was the first polling firm to identify the NDP surge in 2011.

"One of the most accurate polling agencies of the past two federal elections has been Leger."

Quebec separatist, Marcel Leger, is the founder of 2021's most accurate polling agency, Leger Marketing. The company is now operated by his son, Jean-Marc Leger. Despite the Leger family's political affiliation, Leger happened to be one of the most soundly operated and accurate polling firms in the past two federal elections. Days before the 2019 election, Leger pegged Conservative and Liberal support at 33%. Conservatives finished at 34% and Liberals finished at 33%.


The Methods

Each polling agency uses different samples and methods to measure party support in election campaigns. Mainstreet, EKOS and Forum use telephone IVR, or interactive voice response technology. The others use online data and collection, leaving Nanos as the only major polling firm to use traditional telephone data. Ipsos used data collected by both telephone and the internet.

Of the least successful polls from the 2021 election, IVR was the common denominator. Forum Research used IVR, but underscored Liberal support by 3%. EKOS and Mainstreet both relied on IVR and underscored Conservative support by 3% and 5%.

Success and accuracy were highest among those who used online polling and data collection.

Leger, Angus Reid, Ipsos, Abacus and Research Co. relied heavily on online data. Using the average of only those polling agencies directly prior to election day, Conservatives averaged 32% and Liberals averaged 31%. Among the polling firms that used online data by itself, Leger had the largest sample sizes and the most accurate results.

Four days prior to the election, Campaign Research released an online poll with a significantly larger sample size of more than 5,000 participants. However, their data put Conservatives and Liberals at 31% each, 3% less than where Conservatives finished on election day.

There isn't much information available about how each of these companies compile or filter their participants. Some may weigh their samples based on the information they have about the electorate in each province, while others may simply use fewer control methods to separate participants by their partisan leanings. We know that all of the companies have managed to successfully filter non-Canadians and non-likely voters.

Most Canadians use the internet, which makes it a reliable resource for collecting data on partisan affiliations and voting intentions, as long as a voter's age and residency can be confirmed. As with Poletical's two successful CPC leadership polls, IP tracing and location data are reliable resources to filter out multiple votes and non-residents. As well, we know that Poletical's audience is vastly conservative and right-leaning, leaving very little room for liberals to skew the numbers.

Of the one polling firm that relied solely on telephone data, the results failed to accurately measure Conservative Party support. On the eve of the election, Nanos pegged Conservative support at 31% and Liberal support at 32%. While successfully predicting Liberal support to the fraction of a decimal, Nanos missed Conservative support by nearly 3%.


Young Conservatives Don't Answer Telephone Surveys

The data and polling methods suggest that young conservatives and Conservative Party voters are being missed in telephone polls. Despite using a different technology, Forum, Mainstreet and EKOS are still relying mostly on telephone data. As some polls and experts have suggested, Millennials are significantly less likely to answer unusual and unsolicited phone calls than their Gen X and Boomer predecessors.

Conservative support was more accurately predicted in the past two elections by polling agencies that relied solely on online data. Mainstreet, Nanos and EKOS very likely had younger conservatives on their call lists, but they probably didn't answer their phones. This growing reality has likely caused polling numbers to be skewed in the Liberal Party's favour, more so than any kind of political bias.

We don't know enough about young Liberals and whether they are more likely than young Conservatives to answer telephone surveys, but we do know that the data we looked at here suggests, indeed, that it may be the case. By an average margin of 3%, polling firms that rely mostly on telephone data have under-counted Conservative support in several past elections.

We also know that young Liberals and New Democrats are always eager to have their opinions known. Like older conservatives, younger conservatives may have a tendency to keep their opinions to themselves until election day.

© 2021 Poletical