Don’t they use online surveys as well as to reduce this age bias?
Not so much answer the surveys as not being representative of the overall population. I know so many people who don’t answer phone calls, particularly if they don’t know the number of the caller and in many cases, even if they do.
P(votes for trump | answers poll) != P(votes for Trump )
But how different is it?
You are right, that is in the industry. A good poll will reduce selection bias as much as possible in the sense that a respondent’s voting intentions are as close as possible to the general population’s voting intention: P(vote for trump | responded to poll) ~ P(vote for Trump). I guess there are different reasons why the two probabilities may differ.
I could only speculate since I do similar work in insurance, not political surveying. But I’m pretty confident there’s someone loosely similar to me doing that work.
1 Like
I have no idea what they’re doing. The standard approach, I think, is to do a logistic regression based on the features that respondents are willing to give you like age, education, income, etc. Then you have to estimate the distribution of the features for the voting population.
I think the bigger question is are these true?:
P(vote for Trump | Answers surveys) = P(vote for Trump | Doesn’t answer surveys)
and
P(vote for Harris | Answers surveys) = P(vote for Harris | Doesn’t answer surveys)
And is this true?
(P(vote for Trump | Answers surveys) / P(vote for Trump | Doesn’t answer surveys)) = (P(vote for Harris | Answers surveys) / P(vote for Harris | Doesn’t answer surveys))
Let A = responds to surveys.
P(votes for Trump) = P(Votes for Trump | A) P(A) + P(Votes for Trump | not A ) (1-P(A))
You can estimate P(Votes for Trump | A) and P(A), but you don’t have P(Votes for Trump | not A ).
So if P(A) is small then P(Votes for Trump | A) needs to be sufficiently close to P(Votes for Trump | not A) to get an accurate poll.
All this is what Freud would call a “Bayesian Slip”?
1 Like
And that’s what I have doubts on. My expectation is P(votes for Trump| A) is somewhat greater than P(votes for Trump | not A) while I’m not sure that there’s as much difference in P(votes for Harris | A) vs P(votes for Harris | not A).
I also suspect P(A | votes for Trump) != P(A | votes for Harris).
I’d also be curious about:
P(lies about preference | votes for Trump) vs. P(lies about preference | votes for Harris). I suspect that there’s near equal social pressure to claim you’re voting for the candidate opposite who you’d prefer. I’m less sure that there’s equal desire to screw with the polling by pollsters for both candidates. My impression is Trump voters would be more willing to lie in order to mess with the pollsters either being anti-media or anti-privacy invasion.
I don’t do this kind of work, but I think they would begin by estimating p(answers phone | population data). And basically divide by that.
Obviously that doesn’t cover the whole bias, but it should take of some things like age and voter registration.
It kind of sounds like the pollsters are still fumbling around trying to resolve this issue…
This one is a bit disturbing…
The pollsters see no effect, but did find rural individuals more likely to respond than urban. Seems like that could provide a bias towards more rural portions of a voting region which tends to skew rural. Also sample size of 84 compared to a bigger sample, I wonder if they have sufficient statistical power to detect what they’re looking for.
1 Like
In a span of less than 2 weeks, Trump has canceled a rally with the NRA (https://www.ajc.com/politics/trump-nra-rally-in-savannah-canceled/52HMNGS5ZVEUDEWO7Z3YGAG5ZM/), and interviews with 60 minutes, CNBC, and NBC, and held a town hall in which he cut off questions early and swayed to music for 38 minutes instead. (There were medical emergencies at that town hall, but they were both before the abrupt end of the questions). I don’t think that his handlers think that he is there mentally.
I’m quite biased and long have said that Trump is showing his age, but there seems to have been a dramatic decline in the last year or so. My bias is unavoidable but I truly think he’s in his twilight year(s). Probably more than one - probably. His mind is slipping.
It was just less obvious when he was rarely in front of audiences and mostly called into Fox & Friends to ramble until they cut him off.
1 Like
Didn’t Fred Trump suffer from dementia starting at a similar age?
just checked, Fred was officially diagnosed with Alzheimers at 86.
1 Like
With his doctor citing “obvious memory decline in recent years” and “significant memory impairment.”
“Never heard of a Category 5 hurricane before”
Biden went through an accelerated aging between the SOTU address and the first debate. Before then people commented on his age, and he had a few more gaffes, but it could be written off. The stuff I saw on social media in the month or two before the debate…his general movements and actions seemed to show signs of decline beyond his mental acuity.
I think we are in the midst of that same phase for Trump. It’s hard to tell how batshit crazy he is from what he says and posts on TS, but these other data points are all the same pattern. I think he was also recently pulled off stage during an interview because it was “time to go”.
Swaying on stage awkwardly for 38 minutes is not in any way a normal thing to do.
His “eating the dogs and the cats and the pets” line from the debate was not a normal think to say in a national debate regardless of how buried in conspiracy theories you are. Just listen to how he phrased it - it is not even a poorly connected thought. It was the kind of thing you expect to hear when you walk in on your elderly parents at the wrong time.
1 Like
It’s pretty easy to estimate P(A) by just looking at the % of people who actually responded to a survey request. Then after the election you can reasonably estimate P(Votes for Trump | doesn’t respond to surveys ). Do polling companies publish the response rate?
If you had that kind of data for several elections, you could plot the ratio P(Votes republican | responds to survey)/P(Votes republican | doesn’t respond to surveys) as a time series. Of course we’re assuming that people are truthful in the survey and the survey is close enough to the election.