A Musing on Polls

[silverscreenselect]

The presidential polls recently have varied between small leads for Trump and Biden, but what's interesting is that many polls have two sets of results. One set is strict Trump v. Biden, and the other is a five-way race including RFK Jr., Cornel West, and Jill Stein, none of whom are on the ballot in all states, including presumably some of the states in which these polls are being taken. It's pretty clear that West and Stein, when they're on the ballot, aren't likely to get a lot of disgruntled would-be Trump voters. But RFK's attraction is tougher to predict.

But one thing that all these polls leave out is the Libertarian candidate. That won't be chosen until late May at their convention, but they've already said it won't be RFK. The Libertarians are on the ballot in all 50 states, and usually attract voters more attuned to Republicans (except on social issues like abortion) than Democrats. Here in Georgia, one reason that Biden and Jon Ossoff won their races is because of the Libertarian candidates. Jo Jorgenson, a psychology professor at Clemson, got 60,000 votes in a race Biden won by 11,000. And the Libertarian Senate candidate got enough votes to force a runoff, even though David Perdue got more votes in November. Traditionally, Libertarians have gotten more Presidential votes than any other third parties, including the Green Party. So, a five-way poll doesn't really reflect what's going to be on the ballot in November. In states in which one candidate wins by 10 or 20 thousand votes, those votes could make a difference. I'm curious to see what shift there will be in the polls, if any, if and when they include the Libertarian, especially if it's someone with any sort of name recognition.

[jarnon]

Interesting that Perdue finished ahead in November, yet in the runoff, with no Libertarian to take votes away from his opponent, Ossoff won. Maybe other factors influenced the runoff, like Pres. Trump calling his loss a fraud, accusing Georgia government officials and individual poll workers, and asking Congress to throw out Georgia’s electoral votes.

And these are actual results, not polls six months out. I think polls are next to worthless as predictors of the outcome.

(I was going to say polls have some utility in helping candidates decide where to spend time and money. Then I remembered Clinton, who swore by polls, misallocating resources and blowing the election.

[Bob78164]

Interesting that Perdue finished ahead in November, yet in the runoff, with no Libertarian to take votes away from his opponent, Ossoff won. Maybe other factors influenced the runoff, like Pres. Trump calling his loss a fraud, accusing Georgia government officials and individual poll workers, and asking Congress to throw out Georgia’s electoral votes.

And these are actual results, not polls six months out. I think polls are next to worthless as predictors of the outcome.

(I was going to say polls have some utility in helping candidates decide where to spend time and money. Then I remembered Clinton, who swore by polls, misallocating resources and blowing the election.

Nate Silver disagrees with you. And I agree with Nate.

The problem was that most of the mainstream media screwed up by assuming that statewide polling variance would be uncorrelated. That wasn't the case. If the polls are off by a little in Wisconsin, then they're more likely to be off by a little in the same direction in Michigan. When you account for that correlation (as Nate did in 2016), you reach the conclusion that Donny, although behind in the polling, had a realistic chance to win. Hell, a few days before the election, FiveThirtyEight ran an article with four different electoral scenarios. One of them had Donny winning by overperforming the polls by just a bit in the Rust Belt. Sound familiar?

Basically, Donny was down by 1 run going into the bottom of the ninth inning. Sure, you'd rather be ahead, but no one would claim that a team in that situation is toast. --Bob

[jarnon]

I was going to say polls have some utility in helping candidates decide where to spend time and money. Then I remembered Clinton, who swore by polls, misallocating resources and blowing the election.

Nate Silver disagrees with you. And I agree with Nate.

The problem was that most of the mainstream media screwed up by assuming that statewide polling variance would be uncorrelated. That wasn't the case. If the polls are off by a little in Wisconsin, then they're more likely to be off by a little in the same direction in Michigan. When you account for that correlation (as Nate did in 2016), you reach the conclusion that Donny, although behind in the polling, had a realistic chance to win. Hell, a few days before the election, FiveThirtyEight ran an article with four different electoral scenarios. One of them had Donny winning by overperforming the polls by just a bit in the Rust Belt. Sound familiar?

Basically, Donny was down by 1 run going into the bottom of the ninth inning. Sure, you'd rather be ahead, but no one would claim that a team in that situation is toast. --Bob

Maybe my post was unclear. Relying on polls, Clinton neglected Michigan and didn’t campaign at all in Wisconsin. She went to Arizona days before the election. Her last rally was in North Carolina. Smarter tactics could have changed the result in key states.

[Bob78164]

I was going to say polls have some utility in helping candidates decide where to spend time and money. Then I remembered Clinton, who swore by polls, misallocating resources and blowing the election.

Nate Silver disagrees with you. And I agree with Nate.

The problem was that most of the mainstream media screwed up by assuming that statewide polling variance would be uncorrelated. That wasn't the case. If the polls are off by a little in Wisconsin, then they're more likely to be off by a little in the same direction in Michigan. When you account for that correlation (as Nate did in 2016), you reach the conclusion that Donny, although behind in the polling, had a realistic chance to win. Hell, a few days before the election, FiveThirtyEight ran an article with four different electoral scenarios. One of them had Donny winning by overperforming the polls by just a bit in the Rust Belt. Sound familiar?

Basically, Donny was down by 1 run going into the bottom of the ninth inning. Sure, you'd rather be ahead, but no one would claim that a team in that situation is toast. --Bob

Maybe my post was unclear. Relying on polls, Clinton neglected Michigan and didn’t campaign at all in Wisconsin. She went to Arizona days before the election. Her last rally was in North Carolina. Smarter tactics could have changed the result in key states.

The part of your post I was disagreeing with was the part where you say polls are next to worthless in predicting outcomes. Used correctly, I think they're valuable. --Bob

[silverscreenselect]

One thing that trips a lot of people up is that they don't understand how margin of error works. If a poll says that it has a 3% margin of error, that means there is a very high probability (usually 95 to 99% depending on the poll), that the result is accurate to within 3%. But if a poll says Trump leads Biden by 4% in State A with that margin of error, that doesn't mean that it's extremely probable Trump is ahead. The margin of error applies to both figures. So, if the poll says Trump is up 52-48, Trump's total could be anywhere from 55 to 45 and Biden's would correspond. So, it's within the margin of error that the race is a 10-point blowout for Trump. But it's also within the margin of error that Biden is ahead by two points.