Update: Look at Markos’ latest post on the early vote. Simply put, the early vote shows that it is likely that the electorate (again because the early vote is part of the electorate, not a forecast) could include more women voters (including those of color) than is being modeled by almost all the pollsters (who are instead focused on finding “shy” Trump voters, e.g, younger men) . The share of female voters could have course dip as we approach Election Day but it is pretty likely that this increased early vote turnout shows a much greater likelihood to vote by women than currently modeled by pollsters. In other words, the polls ignored this (the Dobbs effect) and are not polling or modeling the likely electorate and we have a reasonable chance to have 2022 like results (where Dems overperform polls and GOP underperforms). Note, in reality the group underperforming (as they have since 2022) are pollsters and the data aggregators and experts.
P.S. I believe the model of the electorate is also now undercounting likely African America and youth turnout but we will see. More to come. As Markos says, keep working. But, also, please, stop the doomerism — it only helps the other side. We are ahead. Let’s go and win this thing.
————————————————————-
If you are like me, you are surrounded by friends and family filled with worry because of the averages.
If you are like me, you check the averages and the “data” experts who are also “bunching” around an election that is “too close to call.”
If you are like me, you, sometimes, find the worry contagious and wonder “is the race really tightening up?”
I’m here to say, to myself and all of you, no the race is not tightening up, if anything, Harris/Walz is pulling away with this race.
Let me tell you why.
We are facing a set of “data elites” (outside of a few) who are pretty smart about math but stupid about modeling human and actual voting behavior. You need to know this and understand the general take of the elites is very, very flawed. These elites are averaging polls that are — by and large — either out and out misleading (like some recent red wave polls”) or modeling the electorate based on electoral outcomes from four years ago.
Polls, in order to be accurate in terms of electoral outcomes, need to poll a representative sample of who is actually going to show up to vote. They need to model the electorate in terms of vote share. The modelers — in talking about this — discuss “voter enthusiasm” as sort of a secondary consideration. But, it is really, the MOST important data point. It is the question about which segments of the electorate are more likely to show up than others. The math to determine who shows up to vote is not the complicated, it is just voters x likelihood to show up at the poll.
If you get “likelihood” wrong, who is going to show up, you — as a pollster — are sampling a group of people who are not representative of who is actually going to vote. This miss is understandable, early on, if you don’t have data about actual actions and intent, are instead reliant on what people say they will do. But, you only do this UNTIL you have actual data on consumer behavior. This is because people are terrible predictors of their own actual future behavior in terms of buying and doing stuff. You do not — as a good analyst of actual consumer/voter behavior — take their word for what they say they will do in the future, you see what they actually do.
To drive this point home, think about what you say you will do in the future — as a single consumer and data point — and what you actually do. Do you find yourself binging that extra show on Netflix even though you had told yourself you were going to bed? Do you find yourself eating ice cream even though you were going to pass on dessert? Do you find yourself having one more beer/drink, even you said you were done? Our actions, as humans, often do not correspond with what we tell ourselves is our rational intent. It is what is mysterious, and beautiful — to me at least — about human behavior.
The thing is that we now have lots of data in terms of actual observable actions — in terms of what voters are doing not saying what they will do — that clearly show that we are getting more of our voters to the polls than the GOP (and what the models used by almost all of the pollsters say is happening). Unfortunately, the vast majority of the polling and self proclaimed elections “data scientists” (lovers of averages) are ignoring and not factoring in this data.
As someone who has — for decades — used data for work, this failure of the modeling and polling elites is mindboggingly incompetent and reflective of the fact that most of this elite are hard scientists and engineers and not people who have worked in social sciences and marketing and spent time trying to forecast actual human behavior.
Imagine if McDonald’s, for example, decided to offer a new sandwich and did qualitative and quantitative research to try and forecast what people think and then offered that sandwich. McDonald’s would, after it starts to sell that sandwich, use actual sales data (and consumer behavior) to decide if they want to really roll that sucker out. If the product manager in charge of the the P&L for that new product, decided to push all in on the sandwich based on the earlier polling data and ignored actual sales and consumer behavior, guess what — they would be fired and very quickly.
Ok, so we now have actual sales data for the election — that is data on actual consumer behavior. We have data on who is registering to vote and now, who has actually voted through early vote numbers. And the so called “data experts” are yelling from the mountaintop that we need to ignore all this data on actual behavior. It is, they proclaim, not relevant!
What Wasserman and others fail — spectacularly — to understand is that what early voting data does do is to show — early on — what voters and demographic groups are showing up to vote early and how that might reflect the likelihood of these types of voters to show up later on and how this could impact the final composition of the actual electorate. In other words, this like — when you offer a new sandwich in test market — seeing who is going to actually buy that new sandwich.
Yes, Wasserman is right, the early vote share is not necessarily predictive of the final outcome in terms of share of vote but it is predictive in terms of actual voter behavior because it is data about people actually voting — not saying how they will vote.
And, the early vote numbers are showing that our people — women, young voters, people of color, etc — are showing in rates that are at higher levels than in previous elections.
The data, from the early vote, is overwhelming, the electorate that is showing up is more Democratic that was modeled earlier.
More women, more people of color, more people from Democratic counties are voting
than has been modeled by the polling and averages experts. Wasserman, Cohn’s and every other pollsters’ theory is that infrequent (or new voters) are going to break for Trump and that — despite the Dems strong performance in 2022 and elsewhere — the general election voters are different and like Donald Trump. They have built their models on who will show up to vote on this basis. And, they have spent their time trying to poll who they think those new voters will be and are now actively ignoring actual data about who is actually showing up to vote and continue to almost solely focus on polls based on models predicated on people’s statement of intent rather than actual behavior.
It’s kind of hilarious that Tom Bonior, who is one of the few sane voices in the room, feels the need to qualify that the sample above is “only 47,000 votes.” That is an huge sample of actual behavior and is so much more important that polls on samples of hundreds of respondents based on models that do not accurately weight the electorate And the actual behavior of voters is — very consistently — showing that they are very, very wrong.
So, I encourage you to stop looking at averages of polls with mostly inaccurate models of the electorate and look at early vote data in terms of who is actually showing up to vote. Even good people, like Adam Carlson, are increasingly irrelevant because they only want to talk about polling. Go to targetearly.targetsmart.com and follow Tom Bonior and others who understand this on Twitter and elsewhere.
Finally, to reassure everyone, don’t worry about the anxiety of the elites too much. None of the voters we need listen, watch or even know who they are. They don’t matter. What matters is that we all keep on pushing and go out and talk to actual voters and win this thing. We are, according to the data that matters, poised to win this thing and win — hopefully — pretty convincingly.
