Transcript for Polling 101: Should You Trust The Polls In 2020? | FiveThirtyEight
The twin sixteen election came as a shock to many news like Times Square has never been quieter right now. The day before the election much of the media seem to be convinced that a Hillary Clinton victory was imminent. You have to say advantage Clinton as we head at these final hours in this big national lead is wide and while the polls indicated that she was the favorite candidate. We all know how that story. Now that we're approaching another. The trump and Biden campaigns are playing on messages of the polling Mena. So you may be thinking how can we ever trust the bulls again do they get 2016 wrong. Well. Only kind undergo that and I inherited explained you what you shouldn't give up on pole. So what happened but the polls in 2016 and what does this mean for 20/20. Before we answer that question let's take a look at what a pull even it is. A poll is not a crystal ball tells you who's gonna win the election. But rather survey taken of a sample of voters lets us guess how population as a whole vote. Conducting a poll ultimately grabbing a handful of jelly beans out of the giants' Jeff if you grab a random Hansel. You can estimate the composition of the whole jar from the ones in your hand and let's take for example this poll and asked college conducted about a week before Election Day sixteen. After calling 940 randomly selected voters maersk found that Clinton led the popular roughly two points. I say roughly because as you might imagine no one can be certain that 940 people speak for the entire country there's always going to be. Fewer people in your sample the bigger your marketers would be larger sample the small I know there. Now this Marist people end up doing pretty well although trump won the Electoral College. Clinton on the popular they just over two percentage points and sixteen about the same margin the poll predicted. The dollar calls from the week leading up to the election didn't do quite as so loathe the polls are all over the place their meaningless right. Wrong when of the key insights we have when doing political analysis at 538. Is that looking at all the polls well always give you a better picture her race and looking at one pole alone. Here are all the polls in the three weeks leading up to election. If we take the average of these polls you can see that together the pool did a pretty good job estimating when the national popular vote would be. Since 2000 national polls have gotten the popular vote right within about four percentage points. They weren't any more often 2016. Have been in previous years. But we don't elect presidents by national popular in this country yet 58 distinct state election. Flanked pulled done to measure how a specific state we'll. Are much more useful in helping us forecast who'll win the Electoral College actually become president. And that is where this story sixteen becomes let twelve mark complicated. In general state level polls are more air crowned the national people's. In any given year most of the state level pulling her tennis everything can. For example in 2012 most locals tend to under estimate Obama's chances. While it may sixteenth scalable ultimately mean to discount Trump's chances. But while the state level polls were off by about 3.4 percentage points on average in 2012. They were five point four percentage points and 26 team. So why I'm where this deal falls more on May sixteenth and asthma imagine their theories. For example there's evidence that more people who told pollsters that they were undecided. And kept casting about for trial. But one big reason comes down how pollsters did something called meeting that responds. Let's return charge arrogantly. There's always the chance that when you grab a handful you randomly pull fewer green jelly beans never in the jars hold. Pollsters deal with this I'm reading their responses. They know what the US population looks like from census data which is supposed to count every single American. At Marist poll we discussed for example had just the responses age gender income race and region to ensure the sample will be representative. Of the demographics of the American population. Basically there aren't enough green jelly beans and your sample the political opinions of the green jelly beans that you do that can be more. If you're trying to estimate the political opinions and attract. But 2016 there's one factor most pollsters didn't take into account only room. King education. And that turned out to be more important indicator of who people would vote for a 2016 in past election. Less educated voters we're much more likely to vote for trial. An analysis of the polling data after the election indicated that they were likely undercounted by the polls. That's one big reason why the polls were more often and sixteen then pass here so heart of the polls meaningless then. Seems like they can't predict we'll win. Well let's not over react here sure the state level polls were of little more often may sixteen and had been in the recent past. But remember this was a very tight race. In Michigan the average of the poll was roughly five percentage points more democratic leaning in the actual result turned out to be. The trump won that state I just want it of 1%. I was also a big part of Biden's selection was so hard to predict. That's the week before the election by thirty eight's forecast gave tribe about three in ten chance of winning. We definitely had an opinion has undergone based on the polling. But even based on these imperfect polls mean new truck victory was a real possibility. Wait so what is song mean for 20/20. Pre question. Well some but not all pollsters have started to wait for education. And polling shows that there may be fewer undecided voters this year but twenty plays that we. But the global pandemic mass protests and an economic crisis a lot of things are going on that could add more uncertainty than usual that's why this year. I'm thirty eight's election forecast this brand new indicate. As usual we're synthesizing national and state level with taking into account the ways that state locals tend to be correlated. And we're considering non polluting factors like how the economy's did. But no. We're also accounting for the additional uncertainty that comes from a year with so much news the next time someone tell you that the polls are all wrong or they can't be trusted send us videos. And remind them that they are in fact useful but imperfect asked me. And telephone poles tend to accurately predict the result in the aggregate but only within plus or minus three percentage points nationally in roughly plus or minus four percentage points state line. And government to mention the margin error and how close the selection was and how all older better look at one. Hey girl again. If you have any questions lending on the conference. If you let this video subscribe to factor yet and you can.
This transcript has been automatically generated and may not be 100% accurate.