Nope, they’re just making shit up to justify why those races were close.
October 29, 2020 5:22 PM   Subscribe

Tired of reading wonky, bland polling analysis? The angry Canadians from Lean Tossup are the cure for what ails you.

Not only do they analyze US election polls, they also analyze Canada and the UK. Last, but not least, they also have a podcast.
posted by reenum (17 comments total) 4 users marked this as a favorite
Texas currently rated as likely Biden pickup? I don't buy it.
posted by tclark at 6:13 PM on October 29, 2020 [3 favorites]

Likely? No. Possible? Yes.
posted by Splunge at 6:15 PM on October 29, 2020 [5 favorites]

I agree it's possible. But their blog entries are pretty clear that they don't mean possible, they really do mean likely.
posted by tclark at 6:28 PM on October 29, 2020 [2 favorites]

You should rethink your resistance to Dems taking Texas: the stats are in Joe's favor there, and there's been a huge demographic shift—still ongoing—in the state that's changing the calculus there. Consider this (and this is before many, many thousands of voters moved into the state in the last two years): last year, the number of legal, eligible Latinx voters was greater than the difference in the last three presidential races. There's a lot to work with there, and there has been a huge registration push from the party this year.

In short, we've heard this before, yes, but this year, Texas really is in play, folks.
posted by ivanthenotsoterrible at 7:11 PM on October 29, 2020 [3 favorites]


In 2016 I would have embraced their level of optimism.

In 2020 I just can't take it. I hope they're right, but I can't take a source this optimistic seriously.
posted by mmoncur at 7:44 PM on October 29, 2020 [18 favorites]

I feel like I've been hearing about the "huge demographic shift" happening in Texas every election year since I last lived in Texas.

(And that was back when dinosaurs ruled the earth.)

I mean, I wouldn't kick a blue Texas out of bed for eating crackers. But I also wouldn't expect to be given the chance.
posted by invincible summer at 7:59 PM on October 29, 2020 [6 favorites]

Texas is a tough nut to crack, and I won’t believe it’s going blue until it actually does. However, a few things:
• The state swung hard in 2018. Beto lost to Cruz by less than three points. There were multiple House district pickups.
• In polling, the suburban collapse in Trump support we saw since 2018? It hasn’t reversed. Also, polls in 2018 underestimated the Dem performance by several points.
• 1.8 million new registered voters. Most of them are in the 10 largest counties.
• Early vote turnout is already over 95% of the *total* 2016 vote. Some counties have already passed 2016.
• Harris County (Houston) should match their 2016 total vote tonight. They have made massive investments in voting infrastructure, and they are running 24 hour voting centers until the end of early voting tomorrow. People in the know say that Harris has to turn out 1.5 million voters for Dems to have a chance in TX. They’re already around 1.3 million. (The @HarrisVotes Twitter account is really good, BTW.)
• Travis County (Austin) is very heavily Dem - and 97% of all eligible voters are registered there. Turnout has been heavy there.
• There is palpable anger at the governor for his order to restrict ballot drop boxes to one per county. Don’t underestimate the backlash from that.
• Last, Dave Wasserman looks at district polls vs state and national ones. They tell a more nuanced story. In 2016 he was sounding the alarm because there were major warning signs for Clinton in swing state district polls. He has access to polls that aren’t released to the public. He was saying last night that the GOP is in legit danger of losing the Texas state house. And he’s bullish on TX - he thinks Biden has a good chance. (He also thinks AZ is likelier to flip than WI, which I’d believe; AZ went Dem for Flake’s old senate seat, SoS, State Superintendent, and we made gains in the legislature.)

Beto O’Rourke has been busy in Texas, not just during his Senate run, but since then as well. He’s been trying to build the foundation to turn Texas blue. Dems need to flip nine seats in the state house to take that chamber and there’s a fairly good chance of that. If they do that, there’s nationwide implications due to how badly the state is gerrymandered. Again, I’ll believe Texas goes blue when it’s actually official. But there’s a better chance than we’ve seen in decades.
posted by azpenguin at 7:59 PM on October 29, 2020 [16 favorites]

this year, Texas really is in play, folks

Maybe this time it won't jerk the ball away!
posted by flabdablet at 8:08 PM on October 29, 2020 [8 favorites]

I came to doubt Texas, but I stayed to make fun of leantossup's web site.

Does it look like a car dealership site on purpose? Just looking at their front page and hood-ornament logo makes me want to test drive something.
posted by invincible summer at 8:26 PM on October 29, 2020 [9 favorites]

Is their optimism right and/or justified? I don't know! I sure hope it is! Did I very much appreciate the bracing, slap in the face tone of said optimism? Absolutely.

Somehow it goes down easier than more upbeat optimism. The brash, vaguely contemptuous tone here gets past my anxiety in a way that Nate Silver's constant hedging and even Dave Wasserman's arcane district polling knowledge can't manage. And given my anxiety has effectively no utility, I'll take anything that gets it to quiet down a little in this final stretch.
posted by yasaman at 8:40 PM on October 29, 2020 [2 favorites]

Does it look like a car dealership site on purpose? Just looking at their front page and hood-ornament logo makes me want to test drive something.

The data visualizations of Canadian elections on their site that I looked at were a hot, incomprehensible mess.

They seem like a few young dudebros who are like, "We have it all figured out! We are disrupting polling!" when they can't even put together a website that is useable.

And it's hard to figure out what they're on about with their unbridled optimism about the American electorate. It's not outside of the realm of possibility that they haven't spent much, if any, time in the U.S.

And my snark might be motivated by hearing their name as a dudebro salutation to an improvised shelter: "Lean-to! S'uuuup!"

That being said, on a personal level I'd like to be cautiously optimistic about what all of the advance turnout portends for November 3.
posted by mandolin conspiracy at 9:10 PM on October 29, 2020 [1 favorite]

azpenguin: "There is palpable anger at the governor for his order to restrict ballot drop boxes to one per county. Don’t underestimate the backlash from that."

I have been hearing this elsewhere - both in Texas ("To me, it's voter suppression", Jess Hardin for The Guardian):
But in many cases, officials are also seeing that voters’ increased anxiety about the safety of their ballots is matched by their determination. “When they see it’s so much harder for them to cast their ballot, they react to that by saying, ‘I’ll show you,’” said the Harris county clerk, Chris Hollins.
and elsewhere, like Detroit (This is voter suppression in 2020 , Nicquel Terry Ellis for USA Today):
"(Voter suppression) does concern me but one thing about the people of Detroit and people of color that I’ve come across... it almost ignites a new fire to make sure that their vote is counted," Rochon said. "The suppression is somewhat backfiring."
I don't for a moment underestimate the effects of all the mechanisms for voter suppression in 2020, but I do find it interesting that people are aware, and talking about it - including the media - and instead of giving up, they're doing everything they can to make sure sure to get their votes cast as irrefutably as they can.

The USPS tampering in particular seems to have gone not entirely according to plan: widespread reporting, starting months ago, led to massive numbers of voters getting their ballots in the mail VERY early, or dropping them in ballot boxes instead of the mail, to try to make sure those votes get counted.

Meanwhile, I sent over 1000 texts to Texas voters this week, and almost as many last week, making sure people vote early and remind their friends to vote.

I won't believe Texas is flippable until I see it certified as flipped, but I'm willing to spend a little very cautious optimism there.

(For an AZ penguin, you sure know a lot about Texas politics! Thanks for sharing with us!)
posted by kristi at 9:26 PM on October 29, 2020 [5 favorites]

Much as I hate to rag on any Worthwhile Canadian Initiative, as near as I can tell, their model is hot garbage, and should be ignored.

So they're not great about talking about their methodology; their methodology page is from April and talks about using national polls, since there aren't any good state polls yet; their current news posts are all hot and bothered about various state polls. So who knows what the hell they are doing.

The good thing is you can download their results, which include the actual random scenarios, which is really helpful -- in a better model, it would be interesting to tease out scenarios; in this one, it helps show how overly simplistic to the point of being incorrect their model is.

So here's a generic guide to how a reasonable model works; then I'll show how their model results differ.

You start with the polls in each state (or some other estimate of the voting result). The second step is to add some uncertainty. Say in state XX, a candidate is +5 in the polls; on average they will finish 5% ahead, but they could finish +4 or +6 or +2 or +8 or even -1 (or +11) under unlikely circumstances. So if you run 10,000 scenarios, the candidate won't always win this state (although they will win the vast majority of the time).

The third step is the hardest part -- these polling errors are correlated. Whatever the reasons that pollsters miss -- demographics under or overcounted, methodological issues, late-breaking decisions, hijinx -- these will generally happen all over the place. One thing that happened last time, for instance, was that education levels had a much stronger overall relationship to the voting, and polls tended to oversample highly educated voters and many pollsters didn't account for this. Which meant that in general, the polls were favourable for Clinton.

But not all states are equally correlated; for example, last time, in part because of the education thing, the polls were particularly high for Clinton in the Midwest (where there are a lot of lower-education white voters); Trump outperformed the polls in OH/MI/WI/IA/MN by over 5%; he outperformed in GA/NC/GA/FL by around 2%, and he didn't outperform the polls at all in CO/NM/AZ/NV. So you need to do some fancy math to represent this in your simulations.

That doesn't mean that everything will happen the same again this time; but it does mean that a model needs to consider the possibilities of correlated errors. If Biden underperforms in NC, he's also likely to underperform in FL and GA which have similar demographics, and that has a bigger effect than just one state. In general, different modelling teams such as the 538 and Economist teams seem to suggest correlations ranging around 0.85 in very similar states (WI and MI), down to somewhere in the 0.3-0.4 range for the least correlated competitive states. (States like Utah and DC have lower correlations, but they don't really matter for the purposes of prediction.)

So Lean Tossup - which is a name I actually love. Step one is the margin in each state. Their state level margins seem unhinged from reality; I looked at 6 major battleground states and compared the average margin from their simulations to the major polling averages. They're about 3% higher than the 538, RCP and NYT polling averages in FL, MI and PA. They're only 1.5% high in NC, but 5% high in AZ. The only state where they are somewhere in the range of anybody else is WI, and I'd guess it's because of today's bonkers +17 poll that may not have been incorporated into Lean Tossup yet. Here's a figure.

Step two is to add randomness. Lean Tossup has a very wide distribution of randomness here; they have a standard deviation of around 4.7% for their states. A 3% polling error like 2016 would happen about half the time with a standard deviation like they're using. In contrast, the Economist has a standard deviation of around 2.2% in their model; which implies last time's polling error was relatively unlikely; one in seven or so. Historically, the polling error is around 2%. So they're adding in way too much uncertainty.

Step three is the correlations. Here, they've done something wildly simplistic; every state has a correlation of 0.8 with every other state (the exceptions are DC, which is somehow negatively correlated, even with Maryland and Virginia, and the various districts within NE and ME, which have 0.99 correlations -- even though Omaha is demographically different than the great plains of western NE). So here's a figure showing their projections for Wisconsin versus the very similar Michigan and the very different Georgia. In contrast, here's the Economist's projections for the same three states. In the Economist, if Biden does well in Wisconsin, he's almost certainly also going to do well in Michigan; he's likely to do well in Georgia, but with much greater uncertainty. This also means that there are scenarios where he could underperform in Wisconsin but still do alright in Georgia. (Note also the axes -- Lean Tossup has a 30 point range in Wisconsin outcomes, the Economist is about half that.)

So basically, they've managed to not do any of the key steps correctly; whatever they're doing is just not reasonable practice. I mean it's fun, but probably there's an octopus who is making predictions or something out there too.
posted by Superilla at 12:01 AM on October 30, 2020 [16 favorites]

If you want someone with even less methodology, there's always the venerable Hedgehog Report, a Republican aggregator. They have it 358-180 for Biden right now.

Grain of salt, etc.
posted by clawsoon at 8:15 AM on October 30, 2020 [1 favorite]

yasaman: Somehow it goes down easier than more upbeat optimism. The brash, vaguely contemptuous tone here gets past my anxiety in a way that Nate Silver's constant hedging and even Dave Wasserman's arcane district polling knowledge can't manage. And given my anxiety has effectively no utility, I'll take anything that gets it to quiet down a little in this final stretch.

I agree with this 100%. I don't take what this site says as gospel, but reading the entries calms down my existential dread over what might happen if Trump wins again.
posted by reenum at 8:51 AM on October 30, 2020

Don't be messed with by Texas.
posted by justsomebodythatyouusedtoknow at 11:42 AM on October 30, 2020

Reading this site is extremely funny with the benefit of hindsight.
posted by Merus at 11:57 PM on November 3, 2020 [1 favorite]

« Older Opium and Bengali comfort food   |   “sentenced the petitioner to a life term, but how... Newer »

This thread has been archived and is closed to new comments