And also because those uncertainties can be quite large, which undermines the credibility of the polling industry.
Also, a nice article on polling errors:
Right, when we restrict “outcome” to mean “winner.” (You and Moriarty both know this; I’m just clarifying it for others).
Excellent read – thanks for posting.
Simon Rosenberg (my emphasis) in today’s column:
[… Watch] Trump admit to paying for fake polls in his Joe Rogan interview a few days ago …
And a reminder - Rs would only be working so hard to make it look like Trump was winning if they believed he was losing.
Tufts University’s Cooperative Election Study was released today. Sample size is 78,247 adults / 48,732 likely voters. Political scientist Brian Schaffer led the study.
(LINK since white-box preview won’t show)
Brian Schaffner @b_schaffner
6hjust released: national estimates from CES pre-election interviews with over 70k american adults
National LVs (Oct. 1 - 25):
Harris: 51%
Trump: 47%Dynamic crosstabs: cooperativeelectionstudy.shi…
Release: sites.tufts.edu/cooperativee…Oct 28, 2024 · 3:14 PM UTC
This is in line with other large-sample pollsters over the last week such as Morning Consult.
51/47 is about where i expect this race to end up.
While there is definitely herding going on, I think it’s also a side-effect of weighting on recalled 2020 vote (which NYT/Sienna isn’t doing, hence they have more movement in their results).
If you assume that the 2024 electorate will match the 2020 one by weighting your responses based on how they claim to have voted in 2020, you are going to get results that are very similar to 2020 but slightly weighted towards Trump (because there will be some number of folks that claim to have voted for Biden but didn’t, and they will be down-weighted).
I think polls have told us everything they can tell us, which is that PA, MI, WI, AZ, NV, NC, GA, and maybe MN will almost certainly be within 3 points one way or the other. It will all come down to turnout and whether the adjustments pollsters made this cycle were correct in modeling the likely voter pool.
My hope is that they went too far, and are understating Harris’ support. My fear is that they are still not capturing the rural masses and young men that are fanatically devoted to Trump.
And while I have been big on defending the science of polling the fact that herding occurs is a big strike against such. Futzing the findings to fit what others are reporting is bad science.
Kind of interesting article about the step-by-step process of the “exit polls” phase.
I’m wondering how accurate these will be, if nothing else due to the “gals voting for Harris but aren’t telling anyone, including their MAGAt boyfriend/husband” factor.
What an idiotic article.
It starts off talking about a red wave that polling aggregators had NOT predicted, goes on about what “average polls” predicted when aggregators do not just average polls.
Anyone who at this point is looking to the aggregators to be reliable oracles or dismissing them as worthless because they are not, is willfully ignorant.
To repeat yet again: historically heavily polled swing state presidential popular vote results are subject to an on average 3.5% systemic error. Not horrible. But if a bunch of likely tipping point states are all polling with 3? A correlated systemic average error one way or the other is a dramatically different electoral college result.
That is not the polling aggregators being wrong; it does not make it a pseudoscience; it is the tool working within its understood limits.
Then polls are absolutely useless for the one thing they’re supposed to do - predict the winner of a race. “The poll said X would win by 1.5, and Y won by 1.7, therefore the poll was accurate” is about as useful as checking the candidate’s horoscopes for Election Day.
BTW, a swing of 3.5% in 2020 = 5.4 million votes. “We were only off by 5 million!” does not inspire confidence.
Does anyone do exit polling of those who have voted early and make that available before Election Day?
They aren’t useless, but in a case where the election is really close (and will be decided by a percentage point or less), they cannot foresee or predict who will come out on top.
A distinction without a difference.
Yes. Those who are thinking that that is the one thing polls are supposed to do will not find that information there except in fairly rare cases where it is a no duh.
They likely want to throw away thermometers as well since a fever or lack of a fever is not a completely reliable measure of a bacterial infection. And cholesterol testing is a pseudoscience since some with normal number have heart attacks and some with high numbers never do.
Of course in each of these cases the tool, including aggregated polls, is just a piece of information in a more complete picture that we use to try to understand what is happening and to predict what might happen.
Aggregated polls have given us a reasonably likely range of outcomes. The medical analogy is to also look at the rest of the patient. That’s where our discussion of what other information tells us comes in, and why some of us are cautiously optimistic, more than the polling forecast models are.
Even within two or three points in critical swing states.
Which brings us back to what started this sidebar to begin with - the observable evidence shows a major enthusiasm gap between the candidates, far more Republicans are supporting Harris than vice versa, and Trump is alienating tens of millions of voters with blatant racism and bigotry, and is projecting an image of weakness, cowardice, and desperation while Harris projects strength and competence.
Harris is going to win and it’ll be over before the end of Tuesday.
Well, yes, but the point I was making is that the margin of victory in 2020 in many of the swing states (Wisconsin (0.6%), Georgia (0.2%), and Pennsylvania (1.2%)) suggest that the outcome in those states are likely to be extremely close in 2024, as well.