In the 2016 elections, the predictions of the opinion polls did not come true: can you trust them this time

“Biden leads Trump in pre-election polls”, “Trump is closing the gap on Biden” and other statements. Can they be trusted, especially after the previous US presidential elections? Explains Air force.

Photo: Shutterstock

On November 8, 2016, many of those who closely followed the voting process in the US presidential election were constantly refreshing The New York Times page with forecasts for the results of the presidential election. Early in the counting process, when results from key states were still pending, opinion polls predicted a landslide victory for Hillary Clinton—more than 80%. Trump's chances were then estimated to be below 20%.

But the predictions made by the authors of the polls did not come true.

Clinton's lead dwindled over the course of the evening. And by eight o'clock in the evening, the probability of Trump's victory was estimated at 95%.

But does this mean that the predictions are wrong and cannot be trusted?

Wrong or not

The most important thing that sociologists predicted was Hillary Clinton's victory by two to three percentage points. And they were not wrong about this: Clinton actually beat Trump in the number of popular votes received by more than 2%. True, she still lost the election, since Trump received the largest number of electoral votes.

When the noise subsided, sociologists decided to find out what went wrong. There were several assumptions that they tried to test.

At first, it was assumed that the error could be due to the fact that some people systematically refuse to answer polls. In particular, poorly educated voters rarely agree to take part in polls.

The second theory was that many of the people who took the polls simply weren't being honest: they might simply be hesitant to admit to outsiders that they were planning to vote for an "unpopular" candidate. This option was possible, especially given the so-called Bradley effect: in 1982, black Democrat Tom Bradley lost to the Republican in the California gubernatorial election, although he was ahead in all the polls. The reason is simple - many interviewees were embarrassed to admit that they did not want to vote for a black candidate.

And the third theory was related to the error of sociologists, who may have incorrectly determined which groups of the population are most likely to go to the polls. No one can predict in advance who will want to take part in the elections and who will not. But sociologists have various models that predict how the electorate will behave on election day. In this matter, even small discrepancies with the real picture of events can lead to significant changes in the results.

On the subject: Professor who has been guessing the election with precision since 1984 named the winner in 2020

Something went wrong

The hypothesis that Trump's voters simply did not want to honestly say who they would vote for was quickly rejected. This theory was easily tested by comparing the results of Internet polls and conventional ones. Research shows that people who are embarrassed to state their real preferences are much more willing to admit them in online polls. Politico and Morning Consult conducted an experiment to test this theory, and did not find any support for it, although they found out that there is a section of the electorate with higher education and high incomes who would more readily admit their preferences in online polls. However, this was a small percentage of voters who could not significantly affect the results. Similar studies were conducted by Yale University and came to the same conclusions.

But other theories have been confirmed to one degree or another. Sociologists found that Clinton's lead in the polls was explained by three main factors. Firstly, a significant part of voters who could not decide until the very end finally decided to vote for Trump - and they made this decision in the last days of the presidential race. Sociologists have not yet learned to predict which way this or that group of undecided voters will lean at the last moment.

Second, the turnout of Trump supporters was higher than anticipated.

And third, Trump's support in the Rust Belt region (which is part of the Midwest and East Coast of the United States) turned out to be much stronger than polls showed. This happened precisely because the survey authors did not pay enough attention to the level of education of the respondents - in 2016, it was this factor that turned out to be important for understanding the election results.

Educated voters are more likely to take part in polls. In a typical US nationwide survey, about 45% of respondents will have at least a bachelor's degree, although nationwide, only 28% of people over 18 have completed college degrees. In 2016, it was voters with higher education who actively supported Clinton; they also willingly participated in polls. And the preferences of people with a lower level of education were not sufficiently reflected in the final sample.

“We know that certain categories of people are more willing to take part in surveys. These are highly educated people, white Americans, older people. These groups are overrepresented in the samples. Historically, this has not had much effect on results. But specifically in 2016, people who had, for example, a high school education (or didn’t even have a high school education) really decided to support Trump en masse—this has never happened before,” explains Pew Research senior methodologist Andrew Mercer.

Conducting surveys and interpreting results

There are two main ways to conduct a survey—by telephone and online. Moreover, the first method has long been considered the gold standard among sociologists. But how do you determine which people to call?

Each state maintains a record of all registered voters with their contact details. These registers are often used when conducting surveys. Thus, it was possible to make samples according to different indicators, for example, by age or education.

Over time, surveys began to be conducted online. And this can cause problems. There is no single registry of emails, which complicates the work. Most often, when conducting surveys this way, companies first use regular mail - sending paper letters to people asking them to take part in an online survey. This method is used, for example, by the Associated Press and Pew Research.

Surveys are also conducted online using another method - when a person simply accidentally clicks on an advertising or any other link, deciding to take part in the survey. Having left his data, a person gets into the database, after which he is periodically invited to take part in surveys.

Such methods have proven to be successful. But here it is very important for companies conducting surveys to follow certain rules on how exactly to interpret the results obtained. Otherwise, the polls are not representative.

The second important question is: how many people do you need to survey to get an accurate result? Pew Research senior methodologist Andrew Mercer notes that surveying a large number of potential voters is by no means a guarantee of success: “The coverage of respondents may be very large, but it is not at all representative, and the results will be biased.”

In the months leading up to the election, news headlines like “Biden widens lead over Trump” and “Poll shows Trump's lead narrowing over Biden” can be misleading. To understand how much you can trust such a survey, experts advise looking at the name of the company that conducted it.

“If you see this or that survey, pay attention to how the question was formulated and what the sample was. It is important that the methodology is spelled out. As a rule, professional companies themselves always want to demonstrate exactly how they worked,” concludes Mercer.

On the subject: Trump or Biden: whose victory is predicted by scientists who guessed the results of the US elections more than once

What will happen now?

In polls released in early June, Joe Biden was 12% ahead of the incumbent. But in three months, Donald Trump was able to close the gap.

Biden now leads Trump by just 7%, according to the latest joint USA Today/Suffolk University poll. Now 50% of respondents are ready to vote for Biden, and 43% for Trump. At the same time, another 7% of respondents have not yet decided. Based on the experience of 2016, it is known that this group of people can have a great influence on the outcome of the vote.

Particular attention is now being drawn to the so-called wavering states. Historically, neither Democrats nor Republicans have a clear majority in these states. In Arizona, one of the swing states with 11 electoral votes, Joe Biden is now more than three percentage points ahead of Donald Trump.

True, one of the Democratic leaders in the region, Larry Bodine, no longer believes in numbers at all: “Polls are an illusion. After 2016, I decided I shouldn't rely on survey results. We need to look at what my Democratic colleagues are facing on the ground. I talk in Democratic circles all the time, and no one talks about polls now. "I think all the promising polls are just giving a false sense of security."

The mistrust in this year's polls is understandable. But there is reason to believe that the mistakes of 2016 were still taken into account. Sociologists pay attention to the fact that now the samples are indeed being adjusted more carefully.

For example, the number of respondents includes the corresponding number of people without higher education. The director of the University of Monmouth, Patrick Murray, notes that if, during the previous elections, experts had adjusted the sample for this indicator, then the Clinton margin would have been only two percentage points, and not four, as the university announced.

The poll companies are also trying this year to predict as much as possible the behavior of those voters who will make a decision at the last moment.

For example, during the last election, Franklin and Marshall University in Pennsylvania stopped polling potential voters 10 days before voting. This was a fatal mistake, since it was in recent days that so many undecided Americans decided to vote for Trump. This year the terms of the survey will be extended.

Program director Terry Madonna notes that this time the percentage of undecided voters is much lower than in 2016: “There are a relatively small number of people who found it difficult to answer. People are really getting into the presidential race this year. Of course, it is important what they decide, but in this particular campaign it is much more important to gain the support of your core electorate.”

Read also on ForumDaily:

Washed floors in the administration: a cleaning lady accidentally won the elections in a Russian village

Biden's headquarters began selling T-shirts with Trump's portrait and the phrase: 'Maybe you shut up?'

NYT received Trump tax returns: he has not paid taxes in years

Facebook has removed the network of Russian accounts due to the threat of interference in the US elections

Do you want more important and interesting news about life in the USA and immigration to America? — support us donate! Also subscribe to our page Facebook. Select the “Priority in display” option and read us first. Also, don't forget to subscribe to our РєР ° РЅР ° Р »РІ Telegram  and Instagram- there is a lot of interesting things there. And join thousands of readers ForumDaily New York — there you will find a lot of interesting and positive information about life in the metropolis.