How not to get overwhelmed by US election polls

DW: New US election polls coming out every day can make it difficult for observers not to feel overwhelmed instead of informed. How can people quickly decide whether a poll is relevant or not?

Ken Goldstein: The answer is not to. Even for a nerd like me, it is sometimes hard to know what to make of an individual poll. So the lesson is to look at the polling averages and not get too happy, too sad, too excited, too calm about any one particular poll. There are three major polling averages here, which are Pollster, Real Clear Politics and Five Thirty Eight. They all use a similar, but slightly different methodology, and they are aggregating all of the polls, both at the state level and national level.

Sam Wang: The quickest way would be to avoid relying on a single poll, but instead look for three polls of the same question, taken around the same time. Then take the middle one, which is called the median. For example, if the three polls show Trump up by two percent, Clinton up by three percent, and Clinton up by seven percent, then the most accurate estimate is that Hillary Clinton is leading by three percent.

There are various kinds of polls being conducted for instance national polls, state polls, phone polls, internet polls and polls of likely or registered voters. Which are some key categories people should look for when reading polls?

Ken Goldstein: I certainly have preferences on methodology and certain polls. But to think that someone who is simply watching that should go into that is probably simply not realistic. So instead focus on the averages and the big picture. With Republicans generally supporting Trump and Democrats generally supporting you should know that all of the volatility you see in the polls is this middle group of voters. So the one thing I am paying attention and you should pay attention to if you want to follow American politics is that group of voters in the middle who really dislike both candidates a lot.

Sam Wang: National polls, if you can get enough of them to take the median, can reveal recent movements in the race. Even better is to look at an aggregate of state polls. At the Princeton Election Consortium we feature a simple snapshot of all recent state polls, viewed through the Electoral College. It is a simple thermometer of where the race is. It is the most transparent snapshot of the race available.

Few people (and pollsters) would have predicted the British Brexit vote and Trump becoming the Republican nominee. How reliable are the US presidential polls these days or put differently should we be prepared for a November surprise?

Ken Goldstein: We have had some notable polling disasters. The Scotland referendum, the Israeli election, the Greek referendum were all cases where the averages of the polls got it wrong. But the polls did not get Trump wrong in the primary. The problem was not the polls, the problem was the people didn't believe the polls. There are huge challenges in doing polls.

The irony is that when polling has never been more difficult we have never had more polls. The way polls could be getting it wrong, is if they get wrong who is going to show up. When polls tend to get it wrong is when they get the shape of the electorate wrong. The big question is whether the young people and the African-Americans will come out to vote against Trump, because they are not voting for Hillary.

Sam Wang: Your question is wrong on all counts. Pre-Brexit polls showed that it was too close to call. People who thought the polls said otherwise were wrong. It was not the polls' fault. Polls also predicted the nomination of Donald Trump. Based on the evidence, I expect polls to be just as accurate in November.

Kenneth Goldstein is a professor of politics at the University of San Francisco. He is also an election night consultant for ABC News and has worked on election night coverage in every US federal election since 1988.

Sam Wang is a professor of molecular biology and neuroscience at Princeton University. He also runs the Princeton Election Consortium, a blog focused on the meta-analysis of election polls.

The phone and email interviews were conducted by Michael Knigge.