US Politics

The evidence of the last two elections showed there were shy Trump voters.

One theory is that there's no reason to be shy about voting for him anymore. He's an ex-president, he's mainstream now and supporting him is a lot more accepted among all demographics than it was 8 years ago.

Some refuse to take part and some lie. As a pollster, you can try to model to take into account people who don't want to take part, but you can't do anything with systematic / non-random lying.

This twitter thread explains it well:

https://x.com/philippilk/status/1853365463629131940?t=Fko3d-hp8sJ4HodbXTfWBw&s=19

One theory is that there's no reason to be shy about voting for him anymore. He's an ex-president, he's mainstream now and supporting him is a lot more accepted among all demographics than it was 8 years ago.

Interesting idea. Maybe - it's plausible.

What the guy says in the thread I linked in the previous post is that there are two reasons:

1. Trump supporters are much less likely to engage with perceived authority / government related discussions like polls.
   
2. 'my wife would kill me' guys.
   

They're both on Twitch now. 9,000 watching Trump, 1,000 watching Kamala.

I'm not on X and won't click on it due to its fuckwit owner.

But what I don't understand is why they would lie at all? What's the point? It's not like polls aren't anonymous. No 0ne is going to find out. Why would you bother lying? Admittedly, I don't understand why anyone would vote for Trump, but seriously, what's to be gained from lying to a pollster?

LOL. Crowd size now, is it?

what's to be gained from lying to a pollster?

Maybe their family can overhear.

In sufficient numbers to skew the results? I find that unlikely.

Looks like Kari Lake has her own methods...

1 Attachment

Are you trying to apply logic to trump voters?

what's to be gained from lying to a pollster?

Approval. People lying to avoid admitting to what they think might be seen as antisocial behaviour is a well known thing in polling, no?

Not to my knowledge. It's an anonymous phone call and completely optional. I just can't see statistically significant numbers of Trump voters agreeing to take part in the poll then lying to avoid some faceless caller thinking they're a bit of a twat. That might be the case for a few, of course, but it seems to me genuinely unlikely that's what a large number of respondents do.

I may be doing them that disservice, yes.

All thos saying "But the bookies!" might want to read this. Not only are malign players gaming the system, but some are planning to use gamed odds as more evidence to back a "Stop the steal" coup.

Not to my knowledge. It's an anonymous phone call and completely optional. I just can't see statistically significant numbers of Trump voters agreeing to take part in the poll then lying to avoid some faceless caller thinking they're a bit of a twat. That might be the case for a few, of course, but it seems to me genuinely unlikely that's what a large number of respondents do.

I wonder about this in households, particularly one in which the man of the house is Trumpy - if the women of the household pick up the phone are they going to say "Harris" in earshot of Mr MAGA?

I really hope that Trumps campaign against women is what undoes him in the polling booth.

In religious households anyone who votes Kamala can be labelled as pro-abortion and in league with the devil. People can't live with each other after that bridge has been crossed.

Joe Rogan has just endorsed Trump. He said Elon Musk "makes what I think is the most compelling case for Trump. For the record, yes, that's an endorsement of Trump." https://news.sky.com/story/us-election-trump-kamala-harris-latest-live-13209921

You have the first part of the problem: Trump supporters not agreeing to take part. But the pollster doesn't know they are trump supporters when they tell them to fuck off. They can correct for demographics but not for this variation within any and every demographic segment.

it seems to me genuinely unlikely that's what a large number of respondents do.

I first came across the phenomenon of poll respondents being unwilling to admit to antisocial behaviour way back in 1992, when it was used to explain how the polls missed John Major's victory.
The argument was that the desire to be liked is a powerful one in human interaction and most of us tend to say things that we think will make the person we're talking to like us, or respect us, a bit more, rather than less. So if you ask people questions where one answer doesn't reflect well on them they're less likely to admit it even if it is true.

Eg, do you drop litter? Do you pick your nose, how often do you drive above the speed limit, have you ever driven after having alcohol? Or the doctors' one: 'how many units do you drink per week'? Which they are alleged to double the answer to, to compensate for systematic lying.

It looks like it pretty much depends on who wins Pennsylvania.

Six swing states:
Nevada, Georgia and North Carolina - look like Trump is ahead
Michigan and Wisconsin - Harris may be ahead.
Pennsylvania - Trump a whisker ahead but well within error margins.

https://www.270towin.com/

Of course it most likely won't be that close (my expectation is that Trump will win most or all of the 6 swing states) but it looks pretty likely that whoever wins Pennsylvania will win overall.

It's 2am and crazy grandpa is still on stage at his rally spewing absolute gibberish

They can correct for demographics but not for this variation within any and every demographic segment.

If this is difficult to account for, how can you be so sure that this time they're under-correcting and not over-correcting or even luckily getting it right? What magic crystal ball do you have that they don't?

It's hilarious how truly captured Rogan is by the cult of Musk, not surprising considering how gullible he is for conspiracy theories. What is amazing though is that he said he avoided joining scientology in the 90s.

That's not what I'm saying. My understanding, based on stuff I've read including the twitter thread I linked above, is not that they're not good at correcting for this, but that they don't try to do it. Because they can't with the range of tools they have available, such as sample weighting. If they did try to correct for it they would not be doing polling, they would be doing forecasting, and that's not what they get paid to do.

I'm not trying to correct for it either, just noting that there was a significant systematic error in the past between what polls said and outcome. In the absence of any reason to believe things are different this time, my default would be to assume a similar systematic error will most likely apply.

Of course some things might be different, such as @Lebowski 's point above, and that could lead to a different polling error and outcome. I'm not close enough to know what will / won't be different. Some things undoubtedly will be, but will they be big enough to make a difference, will they cancel out...? I'm very open to being convinced by a plausible answer to that.

Well, they're defnitely not good at it. Some of the things being tried are crazy and ill-considered (although that particular technique is most likely to exagerrate Trump's support). But I don't even think this particular issue is that significant when it comes to polling inaccuracy. The biggest problem they have is that response rates - people willing to take the phone call, respond to texts and so on - have dropped below one percent of those they try to contact. This means the sample that the pollsters can work with is pretty much worthless, and they try to compensate for that with various dubious models.

When so few people are willing to pick up the phone, it wouldn't help even if they all told the truth.

Just in case someone's reading this and not sure when were finding out about our fates: https://www.bbc.com/news/articles/cde7ng85jwgo