The survey error defining 2018 may not be 2016's. It could be 2017's.

The survey error defining 2018 may not be 2016's. It could be 2017's.

Virginia's presidential election last November looked like Republican gubernatorial candidate Ed Gillespie would take over Democratic leader Ralph Northam. Gillespie, whose track record was decidedly moderate, had won a tough-hitting candidate in the Republican primary and welcomed President Trump's rhetoric on crime and immigration to make up for the difference in his race.

As November 7 approached, the RealClearPolitics average of polls in the state of Northam showed leadership of seven points to about three points. Election day seemed to be a leap.


(Philip Bump / The Washington Post)

It was not It was a blowout. The final survey average Northam would gain about three points; he won with almost nine. In fact, he gained a bigger lead than the average poll ever showed.

The national poll barely missed the presidential election in 2016, with the RCP's average rating believing that Hillary Clinton would win the referendum with three points. She won at two.

Where no surveys took place in 2016 was at the state level. The RCP average had gained two points by Clinton in Pennsylvania, but she narrowly lost. She was three years old in Michigan, but lost a bit. The biggest miss was in Wisconsin, where she rose above six points but was defeated again by Trump.

In other words, the Miss in Virginia was worse than the Miss in Pennsylvania, Michigan, and far and away national. But as a data company of TargetSmart Tom Bonier noticed At that time, the polls had "everything right" – they expected a profit from Northam, and that's what they got despite the margin. The lack of the actual number would be spilled. There was also a precedent for this in 2016: in the poll of the Los Angeles Times-USC Trump had always won the referendum. He did not do it, but he won the presidency – the Times mentioned the "accuracy" of their poll.

After 2016, there was a flood of analysis as to why these government surveys were wrong. On the whole, surveys in general were not uncommonly wrong, as we wrote in June – they were simply wrong in certain places that attract a lot of attention. New York Times' Nate Cohn has focused these errors on three factors in one article over the past year: late movements among undecided voters, failure to weight responses by education (lacking Trump's separation between people with and without college students). Conclusion) and an increase in voter turnout among Trump voters.

This last point is important. To be correct, the pollsters need to assess exactly who is likely to vote. The pool of "likely voters" participating in final polls can be cobbled together in various ways, and co-opting can influence the survey's outcomes. In Virginia, for example, we looked at three different ways of looking at the electorate in 2017, based on self-determined electoral confidence, 2013 federal election participation, or a weighted combination of these two factors.

The most accurate result came from the Pool of Probable Voices, which emphasized self-determined voice security. In other words, the pool that emphasized the enthusiasm.

In 2016, one of the problems faced by government polls was that Trump's voters were stronger than many pollsters expected. In 2017, one of the problems with the polls in Virginia was that Democrats were stronger than many pollsters expected. These Democrats not only elected Northam but a whole host of state-level Democrats and dramatically shifted Virginia's policies to the left.

In recent weeks, the focus has been on the inaccurate polls of 2016, also thanks to Trump and his supporters, who reminded Americans that he and his party had previously been underestimated. All these projections about how the Republicans would lose the house? Well, do you remember when pollsters said Trump would lose Michigan?

These results from Virginia offer the opposite lesson. All these projections of how the Republicans will maintain or expand their leadership in the Senate? Well, remember, pollsters thought the Gubernatorial race in Virginia could end up nearby?

At the time of writing, there are four Senate races where the RCP average margin is less than two percentage points: Arizona, Indiana, Missouri and Nevada. The Republican candidates in Tennessee and Texas have five or six points ahead. If the average of the RCP polls goes down as much as Virginia did last year, all of these races could move to the Democrats and probably give the party control of the Senate.

It depends on the turnout, as they say. It's a throwaway motto that contestants use to increase their turnout and use the media to spot uncertainty in electoral predictions.

It's also a motto that probably has many Pollter nightmares.

Scott Clement contributed to this report.

Leave a comment

Send a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.