Can anyone shed light on the difference between a poll versus a model? I think I heard Ann Selzer say she does not weight their raw data based upon prior elections or education level, for examples. But I think she does weight for some other things to ensure her sample is representative. I'm not real clear on where the line is drawn in terms of what should be described as a poll that has been weighted in some ways versus what should be described as a "model" that is much more likely to be tweaked with assumptions that don't play out. Also, I'm curious because I don't usually dig far enough into them to know. Do the pollsters and/or modelers publish the raw data along with their weighted poll or model?
Since we are throwing completely anecdotal stories about VA out there, I just went and voted. I pick times that aren’t busy, but even at 1030 it was probably the second most crowded I’ve seen it for an election. My district is roughly D+20.
But you can model it against historic outcomes and current polling and get a good idea of where both sides are. In AZ right now, Harris probably needs to be picking off about 12% of Republicans and winning Indies 62 to 38 in order to overcome the early Republican voting lead.
We will see how the eday goes in the swing states. Harris has one of the most advanced GOTV machinery since Obama. Trump has none because he outsourced it to grifters.
In 2020, Biden picked off 9% of the Rs that became crossovers for him. That number could be higher this go-round, as there has been a loud, pro-Harris LDS contingent including some high profile names like former Senator Flake and Mesa Mayor Giles endorsing Harris. And if the Iowa Harris +3 is even close to accurate, there will be spillover not only in the wall states, but Arizona as well, which has a high number of Midwest transplants. As for indies, latest was Harris +8 in Arizona among this group. But winning late deciders by an even larger percentage. Straight polls are exactly that. Models take polls and try and figure out how much each response represents within the state. In a straight poll, a response from Mesa, AZ, and a response from Snowflake, AZ carry the same weight. In a model, the Mesa response carries more weight because there are a lot more residents in Mesa, and therefore, a lot more voters.
I was more pointing our how dominating the brand is there. Where I live (a purple city that went for Biden in 2020) is probably 90% Trump signage everywhere you go. Atlanta was a polar opposite. Just interesting to me.
In the swing states, I wouldn't be surprised to see the Harris crossover be closer to 15% this cycle.
I'm not discounting anything of that. I'm pointing out that early voting is useful in helping determining outcomes. Leaving AZ aside, Jon Ralston is saying that he thinks that Harris is going to win NV in a very tight election but it will have to come from a vote total of 1.3M or greater because of the Republican early voting advantage.
I asked ChatGPT: Summary Polls: Directly reflect current opinions with basic demographic weighting; simpler, and less predictive. Models: Use predictive assumptions and complex weighting to estimate future outcomes, with more room for error if assumptions don’t align with reality. In essence, a poll aims to describe "what people think now," while a model tries to predict "what will happen" based on various assumptions about voter behavior.
Good sign for Trump and Lake considering the early vote advantage the GOP already had. I think Lake could possibly win if this keeps up.
These are excellent points, Andy. You are absolutely right that there is no logical reason why the future must conform to the past. This is Hume’s problem of induction, and it has never received an answer - likely because there isn’t one. All predictions from polling, markets, and models will be subject to error from changing conditions of the universe. Of course, deep learning would not be immune either. By the time these algorithms appeared well trained on data, this data would likely be at least partly out of date. One advantage of markets is that they can subsume all other kinds of knowledge, including polls, models, deep learning networks, and even Hume’s problem of induction itself. Another advantage is that markets are not algorithmic, so they don’t necessarily carry with them all their rules from the past. Or even any. As a result, they may be flawed, but they don’t exactly go out of date. In a way, they don’t even really exist and are more of an emergent phenomenon than an entity. None of this is to say that markets are perfect or even great. Instead mine is a more meager claim that a well constructed market (a truly manipulated market isnt really a market at all) should have advantages over the other methods of predicting. Even so, the predictions of these markets may not rise above your definition of “useless”. Certainly, no thing or no one will be able to perfectly predict the future, but I would like to believe that there is a space between perfect knowledge of the future and utter unpredictability. If so, I think prediction markets can help us map out that space.
Dems are actually ahead of where they were early on ED in 2022. Much better picture later in the day.
In a game of inches, this data is illuminating, to say the least. I remind everyone Trump only needs to flip 6,000~ votes in GA, 5,500~ in AZ and 10,000~ in WI and he wins tonight. He does not need PA, but it would make his path invincible if he wins PA.
Given the GOP's efforts in early voting, it makes sense there would be less GOP voters today, which will lift the Dem % compared to 2022. What's interesting though is that as of that last tweet, I don't see any evidence of massive GOP vote cannibalization.
Maricopa county went 50 to 48 for Biden in 2020. The GOP already have a 118,000~ vote firewall in Maricopa going into today.
Gwinnett County has said they expect to surpass their 2020 vote totals, so doesn't look like a turn out problem there. Hopefully the rest of the Atlanta area follows.
