Trump vs. Harris - Please put all new polls here.

Rather than afoul, I’d use the term divergent. It’s clear these bettors aren’t agreeing with your assessment, but that in itself isn’t evidence of foul play. It could just be that they are seeing something different than you, real or imagined.

 

I never bet on elections, it’s just way too unpredictable ever since Trump entered the race in 2016. I’m fairly confident in a convincing Harris victory as I’ve said since she entered the race, but I wouldn’t be surprised if it’s close or if he even wins somehow.

 

Polls are the latest victims of the politicization of everything.

 

I think useless isn’t quite right. Markets are extremely good at aggregating diffuse knowledge. Prediction markets have likely been the best predictors of presidential election outcomes of any measured. They haven’t been used in the US officially since the 30s, but they hit on 11 of 15 presidential races before that, with the other 4 being extremely tight. They were revived in 1988, predicting that HW Bush would receive 53.2%, which was remarkably exactly what he got. A study in 2008 showed that markets have outperformed polls about 74% of the time.

Given this, our question becomes, if markets are useless, what predictive metric isn’t?

 

It will be interesting to see if inclusion of partisan pollsters helps or hurts these election models.

 

It's 63/37 now on Poly.

 

Billions of dollars being wagered by rich data analysis nerds is a useless measure?

 

I imagine the 11 out of 15 elections prior to the 1930's would have seen better results with the data we have available today and the speed at which said data travels.

 

This will indeed be an interesting question to pursue. Outlier polls are usually “wrong” in themselves, but polling aggregations are often improved by including them. The same could be true of these so called partisan polls.

 

The issue is, I don't see any "respectable" polling that shows Trump is +30 in Ohio, for example. The freakish outliers always tend to favor the Trump opponent, for some odd reason.

 

I was running with my dog past my voting precinct in VA this morning between 7:45 & 8 and there were not any lines. Steady flow of people from what I saw but not a lot of them. I'm in a deep blue district and I think of a lot of people voted early.

 

Indeed there was no scientific polling back then. This knowledge was somehow mined from other sources. These other sources of knowledge, whatever they may have been, likely still exist, which might explain why movements in market odds and poll averages aren’t always in lockstep.

 

Final PA mail in/absentee numbers:

Dems 1,037,783
GOP 618,205

Dem firewall: 419,578 (better than I predicted for Dems)

However, return rate this year...
Dem return rate: 86.4%
GOP return rate: 86.4%

In 2020 it was...
Dem return rate: 87.7%
GOP return rate: 79.4%

Not only were there less ballots sent, which hurt Dems. The ones that were sent, GOP improved their return rate by 7% and Dems fell by 1.3%. If that doesn't seem like much, net improvement for GOP return rate yields +50,109 more votes. Slow down in Dem return rate yields a -15,614 loss of votes. A net difference of 65,723 votes, just in the mail in turnout rate.

 

That is your opinion, based on a study in 2008. Predicting elections has become overly complicated. It has become too complicated for pollsters or any prediction model to properly identify the actual voting population. The voting population changes each election, based on current issues at the time, or perceptions of those issues, and based on the particular election candidates. The unknown Trump in 2016 is vastly different than the known Trump in 2020, which is vastly different than the convicted, desperate Trump in 2024. Also, the 2016 Hillary Clinton candidate was vastly different than the 2020 Joe Biden, which is vastly different than the 2024 Kamala Harris. And the circumstances surrounding all of these candidates has changed dramatically from 2016 to 2020 to 2024.

Also, even if the actual voting population can be identified, which it can't, it has become too difficult to randomly reach that voting population. That is why there is such a statistically large variation on a weekly basis, if not daily.

Finally, the partisan bias funding behind these polls adds too much non-science-based results. All this ad-hoc tweaking to account for poll bias from previous elections is a joke. Depending on the gambling population to represent the general voting population is equally as futile.

The only chance to actually capture this information is through Deep Learning Neural Networks. The rub there is how to properly model and train these networks. The first few attempts at such a model would invariably be off. Over the years, they would slowly improve as methodology to model and train improve. When I say attempts, I don't mean training attempts but rather end-to-end network prediction attempts. Anybody that has even superficial knowledge of Deep Learning understands that training attempts goes into the hundreds of iterations.

By the way, Neural Networks made a huge comeback (from the 70's) starting in 2010. Just noting that to your 2008 antique study reference.

Just my two cents.

 

Im not versed enough to speak to that asymmetry, but I do recall reading an analysis showing that including outliers improved the accuracy of aggregate polling. Of course times change, and if almost all outliers now fall in the same direction, it could certainly bias the aggregate result.

Of course, the is the same argument being made against the partisan polls, where many have suggested that the high number of right-associated polls are swamping the left-associating ones and biasing the aggregate.

 

Dem mail in return rate in Philadelphia county, PA: 79.8%

 

RCP includes so-called outliers. I believe it is helpful to some degree to include them. Most trusted sources aren't going to have outliers that just don't make any sense though. A good pollster will re-poll if they get numbers that don't make sense.

 

For the hundred millionth time:
Straight-up analyses of voting by strictly measuring Reps vs Dems tell us very little, especially when omitting NPAs.

But more importantly, these analyses tell you NOTHING about gender or ethnic participation and crossovers; the most impactful factors in who will win this year.