-538 (sans Nate Silver now)
Well, that’s a shame.
Which of course raises the question of how accurate the model is. I don’t know the answer, but apparently the reason it differs from pure polling is that this far from election day it considers fundamentals like the economy and domestic and foreign policy to be disproportionately important. As election day draws nearer the polls become more heavily weighted. Below is a brief blurb from a Twitter post by someone involved with it, the editorial director of data analytics at 538 and ABC News.
Addressing some confusion about the 538 model:
Broadly speaking there are two types of uncertainty in the model: Uncertainty about how states vote (which we show you) and uncertainty about how they relate to each other (which we don’t show you).
Specifically, we have uncertainty about the correlations between states, informed by (a) a model of the polling data and (b) a model of historical fundamentals data.
Given the disproportionate amount of data and lack of measurement error in election results, we have a much better prediction of historical fundamentals-based correlation than polling correlation.
This influences the “weight” put on each variable. That works like so:
The model has the ability to guess how much polls should revert towards the fundamentals by Election Day. It does this by adding a random variable drawn from a multivariate distribution.
This amount of movement (by design) is not mean-zero across states or simulations. Polls can move up or towards fundamentals — by less when we’re close to the election and more when we’re far away.
Usually this splits the difference between the polls and fundamentals, but the model can push vote shares outside the range of the observed polls and fundamentals if that’s what the model needs to do to make the correlations between states match expectations.
This gives the appearance of us adjusting the forecast to Biden in some close states, eg WI, but it’s because the polls and fundamentals elsewhere indicate he should be doing better than he is in WI — and the model is balancing those correlations between states,
Important! This adjusting will decrease as we get closer to Election Day (and is gone on E-Day) since there is less (no) uncertainty to simulate about how things will move by Nov 5
Finally, this also has effects for the national popular vote. Less uncertainty about the correlation for the fundamentals than the polls increases weight on the fundamentals since the national forecast comes from aggregating the state estimate.
The swing-state polling DSeid cited, plus some specific fundamentals linked below (and yes, I realize the Gaza ceasefire is back on tenterhooks again). There is almost certainly a lag in the 538 predictions, so I don’t expect anything regarding the assassination attempt has yet been reckoned.
There were a few days when the 538 model gave me some hope before I realized that Silver had left and before I took a closer look at their numbers which seem to put a very high weight on their fundamentals. E.g. in PA they have Trump up 3.6 for their adjusted polling average and Biden up 1.1 from fundamentals and the final forecast is Biden +0.5. That seems to imply a weight for fundamentals several times that of polls which seems excessive.
For all I know they are correct to do that, but it can’t be cited as what polling tells us. It’s what one model of predicting voter behavior in which current polling is significantly discounted says. Maybe their belief that the economy actually being good will be what finally rules the roost is true? Silver doesn’t think so. His model is much less sanguine (bad word to use today maybe).
I think part of it is that polls (especially online polls) have been increasingly inaccurate. People have learned to game the system and answer inaccurately out of some combination of embarrassment, trying to please the pollster, or general shit-stirring.
The Economist also has a model which gives Trump a 74% chance of winning which IIRC is similar to Silver. Polymarket gives him 70%. 538’s 49% is a big outlier.
Incidentally there is a nice opportunity for someone to create a model/market aggregator website which averages all the models and betting markets and gives a single number. I don’t think anything like that exists.
Are the models somehow accounting for the apparent “Trump bias” in the primary polls, against which Trump underperformed? Have they explained and somehow removed it?
GIGO, after all. Even the great Nate can’t make chicken salad out of chicken shit.
It’s not really clear that’s true. Polls did just fine in 2022, the last election cycle for which we have data, and in 2018 as well. Public opinion has always been hard to measure, but if you look at all of the data in the article I linked, there doesn’t seem to be any obvious trend toward greater inaccuracy in general, despite some high-profile misses in 2016 and 2020. I suspect pollsters are pretty good at learning from and compensating for their errors in the last cycle, even if new sources of error are constantly being introduced.
Thanks for that! I’ll read it later. I was, indeed operating off the misses in '16 and '20. If they’ve gotten better that’s a good thing.
On the other hand, polls have been very bad in special elections recently, and almost uniformly underestimate the Democratic candidate.
Polling. Click the states, Bllomberg gives WI and MI to Biden.
https://www.realclearpolling.com/elections/president/2024/battleground-states
A post was merged into an existing topic: Polling is broken, right? Or is it the news? Or all of the US?
Bah – all that lead in 538’s model has evaporated. 50% - 49% for Biden plus the same 0.4% of a tie.
Following this model (and others) day-to-day is probably not that valuable of an exercise. Like checking your weight hourly while dieting.
Are you asking – can an ethical pollster put their finger on the results and add a few points for a candidate whose support they have repeatedly underestimated?
No!
He’s asking about the models, not the polls the models use, isn’t he? Or am I completely missing something (which I probably am)?
Much less valuable–it’s checking a 1% change in a model on the percent chance of losing weight by next year.
I saw somewhere that the 538 model uses fundamentals as well as polls, and a bunch of fundamentals (economy, etc.) should be good for Biden. My take would be that this is a very abnormal election and therefore these probably aren’t very relevant, but who knows I guess.
Most other models seem like they are in the 60-80% for Trump, which is absolutely winnable for Biden. It looks like Biden would have to sweep a bunch of states where he’s currently behind but within margin of error. That’s a near exact match of what Trump did in 2016. However it does seem a bit motivated to bump that up to 538’s numbers which would essentially say that Biden would be a toos up to win performing exactly as expected.
While I have posted that Biden should both resign and drop out, the political reasons to do so have declined due to the Trump shooting.
My political reason for changing horses was to get swing/uncertain voters to reconsider a race where, as I read the polls, the GOP had long been ahead. Now that the race has been shaken up in an even bigger and more emotional way, those voters are already reconsidering. In two or three weeks, polls should let us know that result.
By contrast to the shooting, the emotional jolt of Joe Biden dropping out – especially if he dropped out without resigning – would be less. And swing/uncertain probably will not reconsider their vote twice.
Joe should still resign because Harris would be a better president for the next six months. A lesser reason he should do it is to salvage his historical reputation, which – assuming a GOP presidential and congressional sweep – will be low due to stubborn insistence on running for reelection when an octogenarian in cognitive decline. .