The results of the 2016 US election came as a shock to many Americans. Even though Hillary Clinton was leading the polls, Donald Trump still ended up winning the presidency.
As the Election Day approaches this year, it once again looks like the Democratic candidate is in the lead. But after failing to predict the presidency in the last election, can the pollsters be trusted this time?
FiveThirtyEight database journalist Dhrumil Mehta explains why you shouldn't give up on the polls just yet.
A poll is a survey taken by a sample of voters that helps predict the outcome. For example, in a poll taken by Marist College a week before the election in 2016, the 940 pollsters' votes resulted in a 2 point lead for Hillary.
Now, the sample size not so big to determine the actual results, as the smaller the sample size is, the larger the margin of error becomes.
But the reality of Republican backed Trump becoming the President was beyond a lot of peoples' imagination. Trump won the Electoral College and Hillary won the popular vote by just over 2%.
So should we not believe the polls anymore?
Mehta explains why one should not look at just one poll from one year and start distrusting the polls. He said, "One of the key insights we have when doing political analysis at fivethirtyeight, is that looking at all the polls will always give you a better picture of a race than looking at one poll alone."
In all the polls leading up to the Election Day in 2016, Hillary had a lead of close to three points, which is very close to the actual popular vote number.
Mehta said, "Since 2000, national polls has gotten the popular vote within about four percentage points and they weren't any more off in 2016 than they have been in previous years. But we don't elect presidents by the national popular vote in this country. We have 50 distinct state elections. That's why polls done to measure how a specific state will vote are much more useful in helping us forecast who will win the Electoral College and actually become president."
In general, state level polls are more error prone than national level polls. And in 2016, they were on average 5.4 percentage points off.
There are a few reasons behind it. For example, the undecided pollsters ended up voting for Trump because they waited for others to decide. And while so, most did not take into account the educational status of the voter. And that became a grave factor in the 2016 elections in than past elections.
In 2016, the less educated voters most likely voted from Trump. And after election, data counted that they were undercounted in the polls such as the Marist poll where they adjusted the sample size by age, gender, income, race, region to represent the American demographic.
However, 2020 is quite different from the other election years. With the pandemic, mass protests and economic crisis, the uncertainty has increased.
About this year's prediction, Mehta said that they usually do it by synthesizing national and state level polls, taking into account the ways that state level polls tend to be correlated and considering non polling factors such as the economy. But this year they are also considering the amount of news.
That is because the kind and amount of news a voter consumes affects their judgment of choosing a candidate.
This year so far, Joe Biden is on the lead. According to The BBC, 51% Americans will vote for Biden as oppose to 43% Trump supporters.
In the end, a forecast that takes the unusual situation we are in and the effects this has on us into consideration, can be believed to have a close enough outcome.