The Selzer poll, with a margin of error of 3.4, missed the real outcome by 16 points, or by as many as five standard deviations from the true result as revealed on election day.

What are the odds of drawing such a sample by legitimate means? Answer: roughly one time in 3.5 million trials. In other words, given these odds, the results in the Iowa poll likely did not come about by “honest error.”

No one knows now how or why this happened. It is possible that someone in the field made systematic errors in recording survey answers—for example, by mistakenly tallying some prospective Trump votes for Harris. In that case, the error would have occurred due to incompetence on the part of the polling company rather than to outright fraud.

It is more likely that someone deliberately manipulated the sample so that it included too many Democrats, or simply made up the numbers as they came in for the purpose of giving confidence to Harris voters and worry for Trump supporters, or to bring national attention to a poll taken in a state not regarded as competitive.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48338694)

I've posted this before, but these polls are not "polls" at all, and the math of statistics is totally inapplicable.

True polling would be if you selected 10,000 people at random, kidnapped them, scanned their brains, and determined if they would vote and for who. Then based on that sample you could determine the confidence interval of what the majority vote would be.

Modern election polling is people calling 60 people in order to get 1 person who will answer the pollsters question, and that person has almost zero in common with the 59 people who hung up. So pollsters use this data to make assumptions. I believe the better pollsters now ask if they voted and who tehy voted for last time vs who they will vote for this time. So theyre not so much gauging "who did the respondent say they were voting for" but "what % of XYZ votes are switching their votes" and extrapolating based on the last election.

It seems like Selzer was using something closer to true polling, and just ran into a situation where she thought she had a random sample but really the only people happy to talk to her were Kamala voters.

But in any event, none of these polls are anything close to a poll anymore - they're models trying to extrapolate 300 million people based on a small subset of very weird people.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339249)

SC voters were clearly intentionally punishing lindsey graham in 2020, when he polled dead fucking even the whole time and then won by 11.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339385)

it's just sampling error because she didn't want to "project the electorate"

she just called thousands of Iowa landlines, Trumpmos ignored her en masse, the people responding were much more liberal than Iowa as a whole, and her sample was fucked

it was obviously fucked from the beginning, before she published it she knew it was D+2 in a heavily republican state

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339264)

could definitely see her being a Alan Lichtman tier lib boomer wishcasting and convincing herself "WOW YUP D+2 MUST BE A KAMALA WAVE"

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339344)

This. It infers maliciousness when the real explanation is stupidity.

Like if a liberal arts grad conducts a poll with a 5% MOE and misses by 90 points, it's not because they committed fraud. It's because they majored in English.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339354)

her poll was fraud, she knew it, and she did it intentionally because she's a woke Dem crackpot and/or she was getting paid.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339334)

if she did that in 16 and 20 and then inexplicably stopped in 2024 and rawdogged the sample then yeah maybe

it's like Lichtman, he wasn't turning the keys in a bid to demoralize trumpmos, he's just a TDS retard

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339370)

The only way to deal with these problems is to make your poll not a poll, in which case the margin of error is meaningless.

The margin of error is statistical error you'd expect if they're assumptions are right. There's an additional error that their assumptions are just wrong - which is why these "polls" are outside the margin of error like every election now.

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339493)

but this wasn't the same problem. she first refused to release any of her tabs and weights, then when she gave up a bit of it the poll was obviously crazy... and she made up some preposterous story about the entire iowa electorate this year being hyper jazzed 65 year old women who's top issue was abortion. this was an "mistake"

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48339633)

I'm supposed to believe this woman called 50,000 numbers until 800 people picked up?

That's like 2,000 calls a day lol

(http://www.autoadmit.com/thread.php?thread_id=5634384&forum_id=2\u0026mark_id=5304212",#48340004)