Election polls aren't getting any less accurate, but the public trusts them less and less. Here's why.

• Americans' trust in election polls is low, and pollsters' predictions have come under scrutiny when they seem to miss the mark.
• But a look at the history of the polling industry shows they are as accurate as they've always been.
• We turned to experts for the history of polls, Trump's role in public mistrust in them, and what pollsters can do about it.

There is only one perfect political poll - and that is called an election.

But between election cycles, it's the predictions of pollsters that dominate headlines, and often for the wrong reasons.

In 2018, over half of registered voters said - in a poll - that they doubted polls.

But the polls are not becoming less accurate over time. So how did a once-respected industry lose so much trust?

To understand polling today, we need to go back to when it started.

The first political polls weren't as scientific as one would think. Back in the early 20th century, the most reputable forecasting was done by a magazine called The Literary Digest that asked subscribers to mail in sample ballots.

But the magazine's sample wasn't representative of the US population, which led to an embarrassing result in the 1936 election. The Digest had predicted a comfortable win for Republican Alf Landon, although in reality, Democrat Franklin D. Roosevelt cruised to a landslide victory.

Meanwhile, a researcher named George Gallup surveyed a smaller group of people that were representative of the country's demographic. His polls became a hit, and the The Literary Digest would go bankrupt.

The Gallup surveys of the 1930s ushered in a new era for polling - and for voter confidence in it. Back then, Americans were not only willing, but excited to talk to pollsters.

"It used to be such an honor to be selected, to be included as one of the special households who got to speak for thousands of Americans" Jay Leve, CEO of the polling firm SurveyUSA, told Insider. "And that's gotten lost over the decades."

Once telephones were widespread in people's homes, pollsters didn't have to go door to door anymore. But telephone response rates began to fall in the 1990s with the invention of caller ID. Some companies eventually transitioned to online surveys that voters could respond to whenever they wanted. By the early 2000s, there were more than 5,000 polling organizations in the US.

That's when a baseball statistician named Nate Silver analyzed a number of polls and fed them into models that simulated the elections. Silver was able to correctly predict the winner in every state except for one in the 2008 elections. Suddenly, polls were cool again.

Trump's victory changed everything - even though the polls weren't off by much.

Everything changed in 2016, when there was widespread disappointment over pollsters' failure to predict Donald Trump's victory.

But the polls were more accurate than some make them out to be, as Insider's Grace Panetta has reported. They were off by about 4.8 percentage points. That's within half a point of the average error for election polls over the last 50 years.

Part of the problem was that some pollsters didn't adequately weigh in education levels, which ended up underestimating Trump's support, according to an evaluation of the 2016 polls done by the American Association of Public Opinion. For example, college-educated voters were more likely to vote for Clinton, and they were also more likely to respond to surveys in the first place.

Pollsters tried to implement a couple of changes in the 2018 midterm elections, like factoring in educated voters. But a survey conducted by The Hill and HarrisX published right after the midterm elections showed that more than half of registered voters were doubtful about the surveys they hear about in the news media.

When 2020 came around, lack of trust was a big challenge. Trump voters proved even less reluctant to participate, Leve said, in part because of Trump's own role in discrediting and delegitimizing polls that were unfavorable to him.

"If you have been conditioned for four years by tweet after tweet that says, 'That poll that shows me, Donald Trump, trailing is a fake poll,' you may be reluctant to talk to a pollster," Leve said.

In the end, this is his biggest challenge.

"How do we draw a sample that is actually random?" Leve said. "If you decline the opportunity to participate in a survey, you are basically saying, 'I'll let my neighbor speak for me.' Well, that's great - if your neighbor and you agree on every issue."

But the polls did all right. They were off by four points - and that's normal by historical standards.

More than 100 years after the first poll was conducted, the industry has had to reinvent itself all over again.