Davin News Server

From: Josh Rosenbluth <noway@nowhere.com>
Newsgroups: alt.politics.trump,alt.politics.liberalism,alt.politics.democrats,alt.politics.usa.republican,alt.fan.rush-limbaugh,can.politics
Subject: Re: Trump Sues Newspaper Over Election Interference
Date: Fri, 20 Dec 2024 08:13:11 -0800
Organization: A noiseless patient Spider

On 12/20/2024 4:26 AM, NoBody wrote:
> On Thu, 19 Dec 2024 08:31:18 -0800, Josh Rosenbluth
> <noway@nowhere.com> wrote:
> 
>> On 12/19/2024 4:15 AM, NoBody wrote:
>>> On Wed, 18 Dec 2024 08:01:01 -0800, Josh Rosenbluth
>>> <noway@nowhere.com> wrote:
>>>
>>>> On 12/18/2024 4:15 AM, NoBody wrote:
>>>>> On Tue, 17 Dec 2024 11:30:51 -0800, Josh Rosenbluth
>>>>> <noway@nowhere.com> wrote:
>>>>>
>>>>>> On 12/17/2024 10:48 AM, AlleyCat wrote:
>>>>>>>
>>>>>>> Pure election interference, JUST like 2016.
>>>>>>>
>>>>>>> It was election interference, plain and simple. Falsifying polls is the same as telling people they don't need to vote, because your candidate is too
>>>>>>> far behind to win.
>>>>>>
>>>>>> This is just plain, fucking stupid. There is no way the poll was falsified.
>>>>>
>>>>> Did the poll's author ever explain what went wrong with her "poll"
>>>>> that resulted in it being completely out of line with the state's
>>>>> actual results?
>>>>
>>>> It's called an outlier. It's supposed to happen a small portion of the
>>>> time. Pollsters who throw away outliers are engaging in a bad practice
>>>> called herding.
>>>>
>>>> https://www.natesilver.net/p/trust-a-pollster-more-when-it-publishes
>>>
>>> So the answer is "no".
>>
>> She did explain. She chose not to recall weight her poll.
>>
>> The poll asks who they voted for in 2020. In this case, she had a higher
>> percentage of Biden 2020 voters than there actually were. Had she
>> adjusted her sample to match the actual percentage of Biden 2020 voters,
>> she would have had Trump +6 which matches the average of all the other
>> Iowa polls.
>>
> There's no citation direct from her to support this that you've
> provided.

She took a deep dive here: 
https://www.documentcloud.org/documents/25318922-nov-2024-iowa-poll-ann-selzer-review-and-analysis-002/

Theory 4 on Page 10 discusses recall weighting.

> That sounds like a huge oversight.  Wouldn't a pollster,
> upon seeing results like she got note that it was way out of bed with
> reality and find the problem before publishing?

She doesn't know she is wrong until after the election.

>> There were good reasons to recall weight and also not to. On the plus
>> side, in both 2016 and 2020 most pollsters (***) who did not recall
>> weight missed Trump voters because these voters were less likely to
>> agree to take the poll. On the other hand, recall weighting has been
>> lousy when Trump is not on the ballot (see for example the mistakes made
>> by Rasmussen and Trafalgar in both 2018 and 2022).
> 
> Which says nothing to the point.

It's exactly on point as to whether she honestly felt her non-recall 
weighted numbers were correct.

>> (***) Selzer is not "most pollsters." She was the most accurate Iowa
>> pollster in both 2016 and 2020 (the other pollsters underestimated
>> Trump, she stood alone getting his margin of victory correct) even
>> though she did not recall weight. It makes perfect sense for her to use
>> the same methodology that worked in the past when Trump was on the ballot.
> 
> And yet she failed big time here and didn't find the "problem" until
> now?

Again, she can't know there is a problem until after the election.