US Election

Discussions about serious topics, for serious people
Post Reply
User avatar
bolo
Snowbonk
Posts: 453
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: US Election

Post by bolo » Sat Nov 21, 2020 6:34 pm

Bird on a Fire wrote:
Sat Nov 21, 2020 6:15 pm
It'll be interesting trying to hindcast the actual results from poll data with added fudge factors. Maybe assuming that x% of "would not say" should be assigned to Trump/GOP helps.

Maybe old-fashioned polls just don't work any more, and we can do a better job with search data from Google and/or PornHub, truck sales figures and demographics.
Assigning a percentage of "would not say" isn't really possible, as the "would not say" group is on the order of 99% these days.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 6:36 pm

Yeah I've heard that. Remarkable that data like that is still considered the gold standard for predicting votes by politicians and media outlets alike.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 6:37 pm

Trump's lawyers are trying to argue that their utter defeat on the issue of poll watchers in the Pennsylvania supreme court makes the entire electoral system unconstitutional. Their typical attention to detail is present, opening the case by misspelling the name of the Governor of Pennsylvania.

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 6:38 pm

Bird on a Fire wrote:
Sat Nov 21, 2020 6:36 pm
Yeah I've heard that. Remarkable that data like that is still considered the gold standard for predicting votes by politicians and media outlets alike.
Because, despite everything, it's better than any other option.

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 7:01 pm

EACLucifer wrote:
Sat Nov 21, 2020 6:37 pm
Trump's lawyers are trying to argue that their utter defeat on the issue of poll watchers in the Pennsylvania supreme court makes the entire electoral system unconstitutional. Their typical attention to detail is present, opening the case by misspelling the name of the Governor of Pennsylvania.
They are relying on Drake vs Obama for standing arguments. I've not had time to fully digest what how they suggest Drake vs Obama helps them with standing, and I'm not certain how American legal systems handle Ratio Decidendi and Obiter Dicta, but Drake vs Obama was rejected by two courts on a range of standing grounds, which was unsurprising, as its arguments about harm were utterly nuts.

It was a birther case, and the Drake in question was an utterly vile preacher who proudly declared he was praying for god to kill President Obama. Their claim of harm was that as Barack Obama was supposedly not eligible to be president*, a member of the military might get in trouble for following orders, as the orders would have been illegal orders...it was all speculative and ridiculous nonsense

ETA: Right, dug in a little more. Their inclusion of Drake v Obama is because the ruling in that case said that "the “potential loss of an election” was an injury-in-fact sufficient to give a local candidate and Republican party officials standing". The Drake vs Obama ruling went on to explain that as none of the plaintiffs were actually running against Obama, they did not get standing under this principle. Citing Drake vs Obama here is extremely odd, though, as all that line they are relying on is doing is quoting another case, Owen vs Mulligan. The only reason I can think of for citing Drake vs Obama rather than Owen vs Mulligan is some particular familiarity with the former case by the lawyers involved in this case.


*Something they inevitably failed to prove in any way, of course
Last edited by EACLucifer on Sat Nov 21, 2020 7:09 pm, edited 1 time in total.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 7:02 pm

EACLucifer wrote:
Sat Nov 21, 2020 6:38 pm
Bird on a Fire wrote:
Sat Nov 21, 2020 6:36 pm
Yeah I've heard that. Remarkable that data like that is still considered the gold standard for predicting votes by politicians and media outlets alike.
Because, despite everything, it's better than any other option.
I'd be interested to hear more about the other options that have been tried recently and how they compare - do you have a link?
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 7:04 pm

Bird on a Fire wrote:
Sat Nov 21, 2020 7:02 pm
EACLucifer wrote:
Sat Nov 21, 2020 6:38 pm
Bird on a Fire wrote:
Sat Nov 21, 2020 6:36 pm
Yeah I've heard that. Remarkable that data like that is still considered the gold standard for predicting votes by politicians and media outlets alike.
Because, despite everything, it's better than any other option.
I'd be interested to hear more about the other options that have been tried recently and how they compare - do you have a link?
That's my point, though. There isn't really any other option.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 7:08 pm

EACLucifer wrote:
Sat Nov 21, 2020 7:04 pm
Bird on a Fire wrote:
Sat Nov 21, 2020 7:02 pm
EACLucifer wrote:
Sat Nov 21, 2020 6:38 pm


Because, despite everything, it's better than any other option.
I'd be interested to hear more about the other options that have been tried recently and how they compare - do you have a link?
That's my point, though. There isn't really any other option.
I highly doubt that. I expect that polling methods have changed a lot in the past to meet changing social and technological conditions, and can totally change in the future as well.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

User avatar
Woodchopper
After Pie
Posts: 2082
Joined: Sat Oct 12, 2019 9:05 am

Re: US Election

Post by Woodchopper » Sat Nov 21, 2020 8:16 pm

The way to get good polling results is to get a genuine random sample. Do that by picking a household at random, then send a person round to ring the doorbell. If people are out of busy the person returns later, perhaps another day.

The problem is that using that method is enormously more expensive than phone polling, and takes a lot longer.

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 8:23 pm

It's probably worth reminding people that poll performance in this electoral cycle was not that bad, in historical terms. Expectations really are unrealistic after some unusually good polling in 2012 especially, and in 2008.

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sat Nov 21, 2020 8:26 pm

EACLucifer wrote:
Sat Nov 21, 2020 7:01 pm
EACLucifer wrote:
Sat Nov 21, 2020 6:37 pm
Trump's lawyers are trying to argue that their utter defeat on the issue of poll watchers in the Pennsylvania supreme court makes the entire electoral system unconstitutional. Their typical attention to detail is present, opening the case by misspelling the name of the Governor of Pennsylvania.
They are relying on Drake vs Obama for standing arguments. I've not had time to fully digest what how they suggest Drake vs Obama helps them with standing, and I'm not certain how American legal systems handle Ratio Decidendi and Obiter Dicta, but Drake vs Obama was rejected by two courts on a range of standing grounds, which was unsurprising, as its arguments about harm were utterly nuts.

It was a birther case, and the Drake in question was an utterly vile preacher who proudly declared he was praying for god to kill President Obama. Their claim of harm was that as Barack Obama was supposedly not eligible to be president*, a member of the military might get in trouble for following orders, as the orders would have been illegal orders...it was all speculative and ridiculous nonsense

ETA: Right, dug in a little more. Their inclusion of Drake v Obama is because the ruling in that case said that "the “potential loss of an election” was an injury-in-fact sufficient to give a local candidate and Republican party officials standing". The Drake vs Obama ruling went on to explain that as none of the plaintiffs were actually running against Obama, they did not get standing under this principle. Citing Drake vs Obama here is extremely odd, though, as all that line they are relying on is doing is quoting another case, Owen vs Mulligan. The only reason I can think of for citing Drake vs Obama rather than Owen vs Mulligan is some particular familiarity with the former case by the lawyers involved in this case.


*Something they inevitably failed to prove in any way, of course
Pretty much just regurgitating my thoughts in this case as I read it, and more significantly as I read the responses of those who actually know about American federal civil procedure.

Another point to note is that they seek a temporary restraining order to stop certification, while at the same time hacking the foundations out from under that argument by arguing a court could actually reverse the certification anyway. It really, really undermines ones case for an urgent temporary restraining order to prevent irreversible harm if one then goes one to claim the harm is reversible.

User avatar
bolo
Snowbonk
Posts: 453
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: US Election

Post by bolo » Sat Nov 21, 2020 8:54 pm

Woodchopper wrote:
Sat Nov 21, 2020 8:16 pm
The way to get good polling results is to get a genuine random sample. Do that by picking a household at random, then send a person round to ring the doorbell. If people are out of busy the person returns later, perhaps another day.

The problem is that using that method is enormously more expensive than phone polling, and takes a lot longer.
Why do you think you would get fewer people telling you to get lost if you visited them at home rather than calling them on the phone?

monkey
Fuzzable
Posts: 247
Joined: Wed Nov 13, 2019 5:10 pm

Re: US Election

Post by monkey » Sat Nov 21, 2020 9:25 pm

bolo wrote:
Sat Nov 21, 2020 8:54 pm
Woodchopper wrote:
Sat Nov 21, 2020 8:16 pm
The way to get good polling results is to get a genuine random sample. Do that by picking a household at random, then send a person round to ring the doorbell. If people are out of busy the person returns later, perhaps another day.

The problem is that using that method is enormously more expensive than phone polling, and takes a lot longer.
Why do you think you would get fewer people telling you to get lost if you visited them at home rather than calling them on the phone?
It is harder to tell someone to f.ck off to their face than over the phone. I'm sure there are many who would still do that, but it seems a bit less likely to me.

But going door to door in the US and getting a good sample would be the hardest bit. When it gets rural, it gets very spread out and often hard to get to. Especially in the square states and bits with mountains. Enormously more expensive is an understatement.

User avatar
bolo
Snowbonk
Posts: 453
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: US Election

Post by bolo » Sat Nov 21, 2020 9:33 pm

Going door to door is also going to bias your sample. You're likely to miss anybody who lives in an apartment building, or a student dormitory, or a nursing home, or a gated neighborhood, because you won't be allowed in.

And if you're falling back on calling the person to ask for admittance, then you're back to being told no 99% of the time.

User avatar
sTeamTraen
Dorkwood
Posts: 1193
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: US Election

Post by sTeamTraen » Sat Nov 21, 2020 9:38 pm

bolo wrote:
Sat Nov 21, 2020 9:33 pm
Going door to door is also going to bias your sample. You're likely to miss anybody who lives in an apartment building, or a student dormitory, or a nursing home, or a gated neighborhood, because you won't be allowed in.

And if you're falling back on calling the person to ask for admittance, then you're back to being told no 99% of the time.
I think monkey was using the term "going door to door" to describe the general difficulties involved in visiting people at home, based on Woodchopper's idea of selecting the address at random and going round to just that address. I don't think it was meant to imply some attempt to systematically visit every place in a street, which as you note is likely to be fraught with bias.
Something something hammer something something nail

User avatar
bolo
Snowbonk
Posts: 453
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: US Election

Post by bolo » Sat Nov 21, 2020 9:50 pm

Sorry, I wasn't suggesting you visit every house in a street. My point was about attempting to visit random addresses in person, versus trying to call random households on the phone. Just like with phone calls, doorstep visits will fail more often for certain segments of the population.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 9:56 pm

Some run-down from Nate Silver here https://fivethirtyeight.com/features/th ... ty-normal/

It seems that the polls did get the national popular vote fairly accurately, which is nice but the national popular vote doesn't actually matter.

When you start looking at states' races (which do matter), it seems that polls worked in some places and not others, and - which is probably where my impression of unsuccessful polling comes from - a lot of the places with big misses were in the key battleground areas.

I think that's still pretty important, for a couple of reasons:
- key battleground states should have more polling attention. You'd hope that the estimates would be more accurate, not less.
- those are the areas that actually matter for the election outcome.

Key battleground states are also where you actually might need a poll to guess the outcome. I could tell you the California would be blue and Wyoming red without talking to anybody. Places like, Florida, Iowa and Wisconsin are all a bit trickier, so you actually need the polls. What's weird in this election is that the polls made a lot of places look unusually close, when they were in fact as not-close as usual.

In this case, the polls all systematically underestimated Trump, like they did in 2016. The polls in 2018 were much less wrong, so maybe it's a Trump-specific problem.

There's also the issue that the polls seemingly missed entire ethnic groups - Latin voters, Black voters, etc. It's not great for democracy if the political and media classes can't work out how to talk with non-white people.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

monkey
Fuzzable
Posts: 247
Joined: Wed Nov 13, 2019 5:10 pm

Re: US Election

Post by monkey » Sat Nov 21, 2020 9:57 pm

bolo wrote:
Sat Nov 21, 2020 9:50 pm
Sorry, I wasn't suggesting you visit every house in a street. My point was about attempting to visit random addresses in person, versus trying to call random households on the phone. Just like with phone calls, doorstep visits will fail more often for certain segments of the population.
Yeah, I meant going to random addresses, like what was suggested. I did not mean for door to door to mean every door, just going to doors.

I also think you're right that there may be problems in built up places too.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 10:04 pm

Some useful stuff on polling methodology here. https://www.electoral-vote.com/evp2020/ ... .html#cell

Not everyone has a landline (and not every landline is a home address). This is an increasing problem. Relying on landlines will disproportionately miss out younger people, poorer people, students and people less confident in their spoken English.

Almost everyone does have an address. Post is cheaper than visiting in person, but more expensive than robodialling; OTOH with post you can at least reach people who live in apartments and stuff (again, generally younger, poorer, darker-skinned people). It's weird that there's no mention of mail in that FAQ when it seems like a useful source of baseline data at least - anyone know why it's not used? Obviously it's less instantaneous, but you can at least use it to calibrate the voters you miss by other methods.

I'd also like it if polls presented their estimates properly. Instead of just going with their central estimate, which is misleading (e.g. we reckon Biden will get 51.3% of the vote in Mianus, Connecticut) they should present the complete confidence/credible interval (e.g. "we reckon at a 95% level that Biden will get 44.1-57.8% of the vote in Mianus, Connecticut). Even 538 just say "We reckon Biden's gonna win in 89% of our models" rather than a predicted range of EC votes. It's pretty hopeless.

The presentation of estimates of R in covid stuff is similarly misleading most of the time.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 10:07 pm

e.g. assuming the sample outcomes presented are genuinely represented, it looks like 538's predictions at 95% level were for Biden to receive between 258 and 431 votes.

To be fair, they nailed it, but I imagine watching that range blob around for months wouldn't create such a compelling narrative.

(eta just found they do present 80% credible intervals graphically on a slider, with about 275-420 for Biden, so that fits with the above)
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

User avatar
bolo
Snowbonk
Posts: 453
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: US Election

Post by bolo » Sat Nov 21, 2020 10:42 pm

The better pollsters do call cellphones as well as landlines. It's more expensive, because legally you can't autodial cellphones.

But if that extra cost is significant, the cost of visiting in person would be astronomically more so.

I haven't heard of pollsters trying mail. I imagine that a high proportion of mail questionnaires would again just be thrown away, or lost behind a radiator, or returned six weeks after the election. Years ago I used to get consumer surveys by mail (which type of soap do you use? how often do you eat in a restaurant? with ~$10 per completed survey as an incentive to fill them out) so it's not a novel idea. I presume the political pollsters have considered it and rejected it for reasons.

They do online surveys -- very difficult to get anything resembling a random sample.

They recruit panels in advance that agree to be polled periodically so they can track changes over time within the same sample.

I'm sure they do a bunch of other things that I don't know about. The good ones are actually pretty clever and professional. I think there's a danger here of thinking "the results are sh.t so they must be stupid and doing it wrong" when in fact it's a really hard problem.

User avatar
Bird on a Fire
Light of Blast
Posts: 4315
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: US Election

Post by Bird on a Fire » Sat Nov 21, 2020 11:01 pm

bolo wrote:
Sat Nov 21, 2020 10:42 pm
I'm sure they do a bunch of other things that I don't know about. The good ones are actually pretty clever and professional. I think there's a danger here of thinking "the results are sh.t so they must be stupid and doing it wrong" when in fact it's a really hard problem.
Yes, for sure. It's an important problem with a lot of smart people working on it, so I thought that in the aftermath of the election people might have seen some interesting articles - lots of people on the night (including on this thread) were talking about how polling was going to need a big rethink before the next election. It would be interesting to hear more about the specific areas of failures and what could be tried to address them. Perhaps they are waiting for the last .1% of votes to be counted or something.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

User avatar
sTeamTraen
Dorkwood
Posts: 1193
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: US Election

Post by sTeamTraen » Sat Nov 21, 2020 11:03 pm

Bird on a Fire wrote:
Sat Nov 21, 2020 10:07 pm
e.g. assuming the sample outcomes presented are genuinely represented, it looks like 538's predictions at 95% level were for Biden to receive between 258 and 431 votes.

To be fair, they nailed it, but I imagine watching that range blob around for months wouldn't create such a compelling narrative.

(eta just found they do present 80% credible intervals graphically on a slider, with about 275-420 for Biden, so that fits with the above)
To be fair, the number of votes required to tilt the EC from 250 to 400 is probably only a very small percentage of the total. If all they had to do was predict the popular vote, they could just call it now: A Democrat has won 7 out of the last 8 (only Bush in 2004 won it).
Something something hammer something something nail

User avatar
Vertigowooyay
Fuzzable
Posts: 301
Joined: Mon Nov 11, 2019 1:25 pm

Re: US Election

Post by Vertigowooyay » Sat Nov 21, 2020 11:48 pm

Another crushing defeat
Although this case was initiated less than two weeks ago, it has already developed its own tortured procedural history.
This claim, like Frankenstein’s Monster, has been haphazardly stitched together from two distinct theories in an attempt to avoid controlling precedent
This judge is not happy.
Calm yourself Doctor NotTheNineO’ClockNews. We’re men of science. We fear no worldly terrors.

EACLucifer
Catbabel
Posts: 953
Joined: Fri Dec 13, 2019 7:49 am
Location: Behind you

Re: US Election

Post by EACLucifer » Sun Nov 22, 2020 2:19 am

Vertigowooyay wrote:
Sat Nov 21, 2020 11:48 pm
Another crushing defeat
Although this case was initiated less than two weeks ago, it has already developed its own tortured procedural history.
This claim, like Frankenstein’s Monster, has been haphazardly stitched together from two distinct theories in an attempt to avoid controlling precedent
This judge is not happy.
A goddamn idiot who didn't know much law tried to use his courtroom to grandstand. Of course the judge is not happy, and Federal judges aren't known for being the most gruntled bunch to begin with. This is the Giuliani lolsuit, and I'll quote a bit of what Judge Brann had to say, as it is the judicial equivalent of a horsewhipping.
In this action, the Trump Campaign and the Individual Plaintiffs (collectively, the “Plaintiffs”) seek to discard millions of votes legally cast by Pennsylvanians from all corners – from Greene County to Pike County, and everywhere in between. In other words, Plaintiffs ask this Court to disenfranchise almost seven million voters. This Court has been unable to find any case in which a plaintiff has sought such a drastic remedy in the contest of an election, in terms of the sheer volume of votes asked to be invalidated. One might expect that when seeking such a startling outcome, a plaintiff would come formidably armed with compelling legal arguments and factual proof of rampant corruption, such that this Court would have no option but to regrettably grant the proposed injunctive relief despite the impact it would have on such a large group of citizens.

That has not happened. Instead, this Court has been presented with strained legal arguments without merit and speculative accusations, unpled in the operative complaint and unsupported by evidence. In the United States of America, this cannot justify the disenfranchisement of a single voter, let alone all the voters of its sixth most populated state. Our people, laws, and institutions demand more. At bottom, Plaintiffs have failed to meet their burden to state a claim upon which relief may be granted. Therefore, I grant Defendants’ motions and dismiss Plaintiffs’ action with prejudice.
ETA: This is another Federalist Society judge giving them a kicking, so if this is from the judges likely to be friendliest to them...

Post Reply