Original Position,

Their founder was on TV stating that Trump needs to be up by 5 or 6 in PA to counteract all the dem voter fraud. That's just not something a reputable pollster would say. That doesn't prove Trump won't win etc, but I think it tells you all you need to know about Trafalgar.

Unless you are a professional statistician, in which case you should watch this thread to learn more about how they form opinions!

You need to learn how to calculate the standard deviation of a difference. Once you can do that and understand what it means, the answer is that a conclusion cannot be drawn from the data.

Years ago I also worked in aerospace. My answer above is typical of many statements I made to upper management at meetings I would attend.

MM

Yeah, that's outrageous and wrong, but I don't assume it means that their polls are not honest. I've met a lot of sincere cranks who work around politics. Anyway, I agree that they seem like a bad polling firm and their methodology is suspect.

Hi Original:

Along the same lines, I think that Nate Silver is very liberal. But I also believe he takes his job seriously and puts out the best information he can, and there is no question that’s he’s very knowledgeable.

Best wishes,
Mason

While not positive, I think he said 3 or 4.

MM

This is a good example of mindless bad faith both sidesism that has infected politics. My complaint was not that he is conservative. RCP that I've cheerfully linked multiple times was started by a self described "diehard republican" and is defenitely conservative. The problem with that Cahlay guy is that he was blatantly lying about something in his field of expertise. Past election studies have capped voter fraud at around .00002% or something similar. Claiming it's going to be 3 or 4% in PA and knowing which way its is going to go is simply not based in reality. Find an example of Silver doing that and we can discuss.

I know how to calculate that, and I gave you the necessary information to make a conclusion: that the stated confidence interval, which is a function of the standard deviation of each pollster's sample, for each pollster was equal.

I am a scientist by training.

Mason sat down for his nightly indoctrination with clergyman Hannity and got excited by the Trafalgar part of the service. Nothing more, nothing less. Trafalgar is pretty bad of course. Let Mason luxuriate in fever dreams of "new methods" found that confirm the glad tidings of President Trump lol.

Am I right in thinking that a margin of error like +/- 3% refers to the score for an individual candidate, and given in a 2-horse race they are exactly negatively correlated the margin of error in terms of the "winning margin" i.e. A - (100-A) is therefore doubled?

As I said earlier the statistical margin of error is lower if you weight correctly rather than relying on random sampling, but of course there are other sources of error too.

Re: Trafalgar, a lot of pollsters say they do interviews in English and Spanish. I haven't seen Trafalgar say this anywhere but I maybe haven't looked in the right place. That may explain the different average results.

Overseas books have it at something like +190 Trump / -195 Biden

Basically AKo vs T7o.

for those saying that trafalgar is acting as an arm of the republican party, what do you make of the fact that 538 shows them having a mean reverted bias of only 0.9 points?

not saying that they aren't flawed in other ways but there are many outlets on 538 that show similar or greater mean reverted biases that receive a score of B or A.

obviously cahaly looks terrible in the interview but if they were fudging the numbers or choosing methodology to get the answer they want i would expect that we'd see a much bigger discrepancy between their numbers and previous results.


what sites?
hard to imagine lines differ that much.
would be too much of an opportunity for arbitrage.

That was betfair this morning and also my local in Slovakia.

Trump is trading higher now though: +158 Trump / -161 Biden on betfair.

It doesn't really make sense as a conspiracy. If it was a conspiracy, you'd use a talking point that made specific voters more likely to show up and vote.

It can be bias. Which can creep into both data gathering and analysis, either by assuming trends that do not exist or failure to account for trends that do exist. That applies both to subjective bias and methodological bias.

Still, the "networking" approach is an interesting one. We know from social science that one of the strongest predictors for various views is your social circle. Studies that tell us this tend to rely on digital social media data however, not indirect reporting (which of course introduces a very real sampling problem, but such is life). Indirect questions introduce another layer of interpretation, which could be tricky to handle.

Why are polls important to voters? Why should they care about them?

I don't know about voters in general, but I imagine a lot of people here are interested at least in part because they both drive the betting odds and provide information for potential advantage betting opportunities.

Nates being critical of RCP polling averages:

Yeah if you look at a random state like PA, it looks like Trump has made major gains. But it's just an artifact of the way they present the data. By lopping off the final polls from Monmouth etc they are dominated by Trump's best polls. when they should just be taking every outfits last poll, which would show Biden up back where he was a couple weeks ago.

For voters, polls offer useful information about whether or not to vote tactically.

I actually think polls really helped Trump a lot last time in showing Libertarian voters where they might want to vote tactically.

In our completely different system in Slovakia - you need 5% of the vote to get into parliament, otherwise the seats are shared proportionally - it's frequently observable that voters abandon parties still polling around 3% as the election draws closer.

Mostly for entertainment purposes.

It’s possible there is a bias on their part. It’s possible it’s even intentional. But they have no more of a temptation to intentionally misreport than last time around and they just weren’t off by that much last time.

Our best estimate would be just to look at their previous mean reverted bias and apply it to their current poll figures. And similarly we’d want to do the same for pollsters that chronically overstate D support, which there are many of to the same degree that trafalgar has overstated R support.

I think this is what the betting lines reflect in most cases, and the biggest error the lines commit is the extent to which they over/underestimate sampling error from limited data. Ie: does being off by 1% represent a 40% or of the trailing party to win? 33%. Depends on the standard deviation of polls / how big the margin of error is.


As for the incentive for conspiracy conjuring - I think this is more to get publicity than anything. Kanye vote spoiling for trump in Pennsylvania? He needs only to look at his own polls where “someone else” captures only 2% of the vote of which Kanye is a small subset, and not all of his votes are coming from trump obviously. He likely doesn’t believe it. This is just an “any publicity is good publicity” thing.

Tell me there is a flaw in this poll:

Elections themselves are a very small sample size. I mean think how many games it could take to tell a .300/.500/.800 hitter from a .200/.300/.350 hitter and one is an MVP level player and the other shouldn't be in the majors. But in a 7 game series they can easily be indistinguishable. We have probably less election data than that and we are specifically post selecting Trafalgar for supposed over performance. Of course if we had more data (ie they were around in 2012 like the unskewed guy) nobody would be talking about them now.

The sample size is small for an individual state in an individual poll. Over all states the sample (typically around 500-1000 per poll) has a margin of error where that’s quite small. The nature of the “bias” is what we’re really arguing here.


Interesting to see how these pollsters will be looking when the dust settles. Looking better
For trafalgar than silver so far.

Well, if the person who designed the poll is a scientist by training I guess it must be good.

Pollsters Taking A Beating ...

https://thehill.com/opinion/campaign...akes-a-beating

Pollsters may well be having a problem getting honest answers across the board.

It is not so much that people are trying to sabotage the polls with outright lying, it is more that people have a tendency to give people the information that they want to hear.

The legacy media has engaged in a crusade against Trump. Not only that, they oppose anything and everything he does and says. Trump is the most socially unacceptable President in living memory. Respondents may just be telling the pollster what they think the pollster wants to hear.

This squares with an opinion expressed earlier by my (very conservative) girlfriend, (i.e. that Trump voters were not telling pollsters they intended to vote for Trump), and this tendency skews the poll's accuracy. (I think the term of art for this tendency is "Trump shy" voters ...)

I recall asking Mason how pollsters would detect this tendency and how to measure (and adjust) for it? (I wonder if it's even possible to detect such a tendency?) I suspect yesterday's election may be another example of "The Bradley Effect"

https://en.wikipedia.org/wiki/Bradley_effect