Friday, November 1, 2024

If you believe MU Law Poll, the race in close in Wis. But it doesn't seem like 2024's reality

As we get into the last week of this election cycle and numerous polls are thrown into the atmosphere, and our media uncritically reports those numbers to come up with the "state of play", this article from Vanderbilt University political scientist Josh Clinton rang true to me. He notes that "poll results depend on pollster choices as much as voters’ decisions."

We need to remember that pollsters don't just take every response and produce a number, but also take demographic data and create an electorate that they think will match who casts the votes. And that is something that is harder than ever when fewer people pick up phone calls from weird numbers. One example that Josh Clinton uses is how pollsters guess as to how partisan the electorate is.
For partisanship, pollsters often rely on benchmarks such as the Pew Research Center’s National Public Opinion Research Sample, which suggests that the country is evenly split – 33% Democrat, 32% Republican, and 35% independent – or Gallup’s tracking survey, which suggests that 28% are Democrats, 31% are Republicans and 41% are independents. If I adjust the raw data by both the demographics of the 2020 electorate and these party identification benchmarks, Harris’ margin is greatly reduced relative to the raw data and demographics alone:

By the way, right after Clinton wrote this article, Gallup gave an update on its tracking survey, and it now says 32% Dem, 29% Republican, and 37% Independent. Needless to say, if a poll's makeup went from R+3 to D+3, it would likely result in a 5-6 point swing in the polls. But I bet most pollsters haven't made a change like that to their models this week.

Clinton also mentions that pollsters need to decide how likely it is that someone will vote, and how much should an enthusiasm gap play into that analysis.
This poll asked respondents to rate their likelihood of voting on a 1-10 scale, where 1 means they definitely will not vote and 10 means they definitely will. Here is what happens if I weight the sample so that each respondent counts “more” if they reported a higher number on that scale.

First, because Democrats are slightly more enthusiastic than Republicans who are themselves slightly more enthusiastic than independents in this data, adding the likely voter weight moves the margin ever so slightly towards Harris:

Pollsters also need to look at age and gender, especially in an election like this one, where there is likely to be a wide variation between various demographic groups. A similar choice needs to be made about new voters, both in identifying if they are registered voters, and in adding in how much of the electorate should be those voters.

Which takes us to this week's Marquette Law School Poll of Wisconsin, the alleged "gold standard" for this state. I made a wistful prediction about what they'd say an hour before the poll came out.

Oh, but I was wrong. MU slanted their poll even harder toward the GOP.

Yet IN THE SAME POLL, they had this response.

So Democrats have more enthusiasm than Republicans in Wisconsin, and the Democrats got more enthusiastic from September's poll, while Republicans did not. Yet the MU Law Poll's electorate got MORE Republican from the already-off R+3 in September? How does that work?

The reason I am angry at this is that it is a conscious choice by Charles Franklin and company to make the electorate more Republican. Likewise, the gender breakdown of the poll is 50.1% women, 49.9% men. And indeed in 2020, the gender breakdown in adjusted exit polls was 50-50 in Wisconsin (while being 52-48 nationwide, which is...interesting), and that seems to be where a R +5 assumption could come from for the state's electorate. But in 2022 (after the GOP's election denials on January 6th and after the Dobbs decision outlawed abortion in Wisconsin), the exit polls had a Wisconsin electorate that was 53-47 women, and dead even between Democrats and Republicans.

It makes you wonder why Marquette decided to revert to these 2020 numbers and ignore what happened 2 years ago, or ignore anything that's happened since then. There's a 33-point gender gap for the presidential race in the Marquette Poll between men and women (Trump +15 with men, Harris +18 with women). If we even split the difference between 2020 and 2022, and made it 51.5-48.5 women, that would add 1 point to Harris' lead. If we made the electorate R +2.5 instead of R+5, that adds 2 points to Harris' lead. And we'd have a race that would be a 3-4 point lead for Harris instead of 1, which would not make the race seem like a toss-up to the press.

Which leads to this excellent article from earlier this month by Denny Carter in the Bad Faith Times, where he reminds us to "watch the game", and not the polls or the pundits.
Peel your eyes away from the computer screen showing a two-point swing toward Trump among men in Michigan with some college education who were born under a full moon between October and December and you might encounter a different reality – one in which one candidate is behaving like the favorite and the other is thrashing around like me when I get a calf cramp in the middle of the night.

One candidate, after securing and firing up the base with some of the best campaigning of my lifetime, seems to have pivoted toward undecided voters and possibly some Republicans who are so repulsed by Trump and his open embrace of fascism that they might consider the unthinkable: Casting a vote for a Democrat (I don’t love Harris campaigning with Liz Cheney but I can report the normies love the shit out of it). We get snippets of data showing a big old chunk of voters in swing states say they support Trump’s policies but can’t stand his demeanor, and are therefore undecided or wasting their vote on a third-party candidate.....

We have one candidate selling gold watches and cryptocurrency and sneakers and shit I’m sure I haven’t even seen in a last-ditch effort to squeeze the grift dry, to drain his frothing congregation of every last dime before the whole thing collapses on itself. The other candidate is doing her job as vice president while hustling to create a Coalition of Normies that can deliver the final blow to this particular form of the fascist menace.

One candidate has packed stadiums full of cheering throngs; the other is conducting a gaslighting campaign about his sparse rallies, saying the people filing out of the venue in the middle of the rally aren't actually leaving the rally. But they are, because they are Americans, and they're bored in Season Nine of this fascist clown show.

Which of these candidates seems to be in command? Don’t ask the computer. Be honest with yourself about what you see. Watch the damn game.
RIGHT. Go look at the crowds at the dueling rallies in Milwaukee tonight, look at how the candidates are acting, and that might tell you more than the absurd amount of polls that have been published about this race.

PS - Now we have confirmation from the New York Times' Nate ("I'm a Polling Expert, Because That's the Job Description") Cohn that many organizations are intentionally suppressing great polls for Kamala Harris, and for a very lame reason.

OH REALLY? So we see the outliers that help Trump, but we don't see the outliers in favor of Harris? Well then we're making our predictions on incomplete data, aren't we?

Let's not pretend it's in the bag, and a lot of work is left to do (after all, Dems need to win by 4-5 statewide in Wisconsin to flip the State Assembly, and Rebecca Cooke needs to beat Small-D Van Orden). But I have a suspicion that things are better than the toss-up that the polls and the pundits would indicate.
Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)