Twitter | Search | |
Nick Gourevitch
One of the polling lessons I learned from conducting polling almost up until Election Day in both the 2016 general and the 2017 VA Gov race is that “polling error” is partly late movement (partly!!). 1/?
Reply Retweet Like More
Nick Gourevitch Oct 31
Replying to @nickgourevitch
There were obviously some methodological issues with 2016 polls around education weighting and urbanity but there was also late movement. In 2017, a lot of those were addressed and we still saw late movement towards Ds.
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @FiveThirtyEight
This is very difficult to untangle if you are an outsider like judging polling error. Especially because some polls do have error and methodological flaws.
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @nickgourevitch
But I’m really not a fan of judging polls simply by the margin of final result to margin of Election Day. You can do very rigorous poll 2 weeks out and it can look wrong if the race shifts. You can also do a garbage poll 2 weeks out that looks right if the race shifts.
Reply Retweet Like
Jacob Segal Oct 31
Replying to @nickgourevitch
then how do you judge polls?
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @nickgourevitch
Campaigns spend the most resources in the final week(s). Plenty of voters only tune in at the end. The expectation that a poll done 2-3 weeks out needs to predict the result to be “right” is bad. This is especially true in lower info races (House, state leg, ballot proposals).
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @nickgourevitch
As campaign pollsters, we almost never do polls in the final days because most decisions have been made and it’s mostly curiosity at that point. But every cycle we do a few and every cycle those are more accurate than the ones we did weeks prior. Imagine that.
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @nickgourevitch
Anyway, late movement is a thing. It’s real. We’ve done callbacks on pre-election calls to confirm it. All these races aren’t settled. 6 more days. The end.
Reply Retweet Like
Jason Pipkin Oct 31
Replying to @nickgourevitch
I suppose in states with heavy EV, you might be able to get away with testing polling accuracy at earlier time points by how well it captures the "already voted" margin. but even then EV electorate composition changes over time
Reply Retweet Like
Gabe The Centrist Oct 31
Replying to @nickgourevitch
Honest question, wasn’t it only the polls in the last week that showed Trump within striking distance in PA and MI? And the single MI poll with Trump winning was by an R pollster? Why MI particularly was so surprising.
Reply Retweet Like
some dude Oct 31
Replying to @nickgourevitch
In fact, isn't late movement in a uniform direction one of the indicators of a wave? Happened in 2010 and 2014 in the dozens and dozens of surveys I saw in those cycles.
Reply Retweet Like
Eric Heggie Oct 31
Replying to @nickgourevitch
Thanks for this update Nick.
Reply Retweet Like
TKN-NOLA Oct 31
Replying to @nickgourevitch
Could you please look at this data 👇🏽 & explain how 24 of 28 states shifted from blue to red (ExitPolls vs VC) in 2016. Odds 1 in 13.1k 🤷🏻‍♀️ Silver had HRC at 80+% still around 8pm election night. Would 🇺🇸 question results like this in elections abroad?
Reply Retweet Like
Jason Elan Oct 31
Replying to @nickgourevitch
Reply Retweet Like
JRMarlow Oct 31
So what’s the point of a poll three weeks out if not to predict the result? How can we asses the quality of polls if we can’t compare them to the result?
Reply Retweet Like
Jonathan Robinson Oct 31
Replying to @nickgourevitch @Civiqs
(just an ex.) disagrees on late movement in VA
Reply Retweet Like
Nick Gourevitch Oct 31
Replying to @jon_m_rob @Civiqs
Yep. People will debate this until their dying day. We saw differently in VA. I used to be a “nothing changes” person before 2016 but now I’m a “things are pretty darn stable but not immovable” person
Reply Retweet Like
Patrick Ruffini Oct 31
I think you can probably chalk up the (meaningless!) 1 point national polling error in 2016 to normal third party overestimate and Johnson voters breaking 2-1 for Trump.
Reply Retweet Like
Jonathan Robinson Oct 31
Replying to @nickgourevitch @Civiqs
I'm like 80% nothing changes, but I hear ya!
Reply Retweet Like
Patrick Ruffini Oct 31
At least that's what CCES shows.
Reply Retweet Like