Election polls are similar to a horse race, and while people like to watch them to pick a winner, root for the underdog or just figure out what’s happening, they’re not always accurate depending on the pollster, maintains Livianna Tossutti, a Political Science professor at Brock University. She was addressing the topic of How to Decipher Polls as a part of the Lifelong Learning Centre’s Speaker Series in St. Catharines, Wednesday.
“People like horse races. Politically, they want to know who’s ahead or behind,” she said. Election polls, which date back to the 1940s in Canada, can positively and negatively influence people’s voting behaviour, which is why they're important. On a positive note --polls keep people engaged in the electoral and democratic process.
“It also helps the media perform its watchdog function,” she added.
On a negative spectrum, a person’s chosen candidate not doing well in the polls may dissuade them from voting. The polling industry expanded “when polls became news,” said the three-time teaching award-winning recipient.
While there have been a few high-profile mistakes in polling, “They often get it right,” said Tossutti, pointing to the electoral races in 2011 and 2015 in Canada, which resulted in Nanos, Harris-Decimas, Angus Reid and Ipsos Reid predicting the election results with a few percentage points different from the actual results.
With a slide presentation, she gave a brief overview of what the polls showed regarding the Brexit vote in England and the last U.S. presidential election.
The methodology of the poll, or how it’s designed, significantly affects its accuracy, she said, unravelling the mystery behind the numbers. For more accurate results, polls need to be random, which means they're strategically chosen to represent certain characteristics of populations or groups that are diverse in culture, religion, education and class. However, many polls conducted can be based on data that is not random or representative, because the participants are often anyone “who answers the phone or completes a survey,” which is not random. If polls are not randomly conducted, people should be concerned with how the information is being used, she warned. This type of sampling also implies bias.
Focus groups using volunteers are often used to remove bias in surveys, but “they’re a different breed,” than non-volunteers and not representative of the general population, said Tossutti.
Minorities, who have language barriers, are especially unrepresented in the polls and polling results.
One of her biggest concerns is how pollsters determine their results given the number of people who don’t answer the question or are “undecided.” How these answers are allocated can skew the averages and numbers in a race. In Canada, the average response rate to surveys is seven to 10 per cent, she said.
“So how do pollsters try to make the math adjustments when they don’t get the response? How do they get the response rate up?” she asked. Tossutti believes so far, Americans have been better at figuring this out, than Canadians.
Polls can be incorrect in their estimations for several reasons, including the fact that people lie when questioned for various psychological reasons, such as wanting to give what they think is the right or agreeable response to the person asking the question. Even those who answer honestly may not end up voting, which “renders the poll useless,” she added. The expectation for voter turnout can vary.
“We know that people who are in higher income brackets and are more educated are more likely to vote, so we should, perhaps, weight them lower.”
From time to time, statistics show youths have voted in higher numbers than expected, and seniors don’t always show up at the polls, which can affect the end result come election day.
TYPES OF POLLS
Currently, polls are most often conducted by phone interviews, because it's one of the most economical methods. In the past, there were pollsters who interviewed people in their homes, but that became too costly and many people will not open their doors to strangers anymore. After 2005, cell phones became the norm, land lines declined, and telemarketing increased, which posed a problem for pollsters attempting to gather data by phone. Eventually, phone interviews decreased. The Internet was supposed to solve this problem.
Automated polling techniques like Interactive Voice Response and Artificial Intelligence are the latest methods of gathering public information. Tossutti noted many people don’t like automated interactive voice, or talking to a machine, so the response is not always noteworthy. However, people who do participate in these types of surveys tend to “tell machines exactly what they think,” and be more truthful, expressing answers that may be unpopular or not politically correct. She refers to this fringe group as the “outlying populations.”
The latest trend in polling is called artificial intelligence, where polling analysts peruse the Internet for social media activity, opinions, polling history and news media reports before applying statistical equations. This can include Internet surveys, which are often completed by “outraged people” or those who tend to complain, she said. However, certain analysts have been able to adjust for “the noise and partisanship” in the results.
Tossutti finds artificial intelligence problematic, due to mistruths and fake news printed online. Also, the respondents are not random. A significant number of people don’t use social media.
She also cautioned against any surveys that only have a few hundred respondents, not large enough to be accurate. If the methodology is correct, and the proper time is taken to conduct a random survey, 1,200 people is a large enough sampling. Polls conducted by a certain party or company that isn’t “independent” will reflect a large degree of vested interest and bias. Although, she added, “I don’t think there is such a thing as an unbiased survey.”
Certain companies conduct surveys which combine political or social issues with a marketing need. For example: a pollster may ask about a policy issue, and then add a question like, “What type of toothpaste do you use?” She said it was important to pay attention to the sequence of questions, and how they're asked to determine its bias.
THE TRUST ISSUE
An audience member asked if we should trust surveys conducted by a political party. Tossutti replied many political parties hire independent or academic pollsters, which are more trustworthy.
While she wouldn’t state who her favourite polling companies are, she said Poll Aggregation companies tend to be most accurate, because they combine many polls over a three-day period, for example, and then tabulate the results. Since these polls rely on many sources, they tend to contain more accurate results. She cited Grenier’s CBC’s Poll Tracker as a good example. Knowing the track record or success rate of the polling company in determining results is key.
THE ERROR GAP?
Most polls have a margin of error of about two or three per cent on a sample size of 1,000 people, so when a political leader and the runner-up are close in a race, the results can be deceiving. On a smaller scale, local ridings have smaller sample sizes, so races are even more difficult to predict when the lead candidates are both performing well.
Readers should also note who paid for the poll. This information should be published with the poll's results, and while in most cases they do, polls don’t always indicate how they calculated or factored-in the “undecided voters” or the “don’t know” voters, as well as how they calculated the margin of error. In Eastern Canada, the sample size is only 300 people with the margin of error at five per cent. This can result in a significant shift in the numbers, especially once the undecided voters show up at the polls.
Another audience member said she believed voting was supposed to be private and confidential, fearing that once you participated in a survey, your privacy was gone. Under the Canada Elections Act, Tossutti asserted pollsters from political parties are required to remove any “identifiers,” however; she knew of a party or a case where this did not happen. Despite that fact, Tossutti said she did not want to dissuade anyone from participating in the polls, alluding to what she stated earlier, that polls support the democratic process.
