Nate Silver

Nate Silver @NateSilver538

I know I've made this point before, but one of the reasons that there's high uncertainty about the election outcome is that there's high uncertainty about what COVID will look like by November. https://t.co/Qxu8rT6ZMz https://t.co/XfwgqFaSPX — PolitiTweet.org

Nate Silver @NateSilver538

RT @micahcohen: This involved a massive amount of data analysis, reporting, visual journalism and more by a ton of very skilled people. R… — PolitiTweet.org

Nate Silver @NateSilver538

RT @juruwolfe: So many people across @abcnews & @FiveThirtyEight worked tirelessly for months to report out how Black communities had less… — PolitiTweet.org

Nate Silver @NateSilver538

Excluding a day with a data anomaly, this is the first time back over 1,000 deaths since May 29. Unfortunately, can't say it's a huge surprise given the U.S.'s poor trajectory; we'd nearly hit that number several times and Tuesdays tend to feature heavy reporting. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 1,029 Yesterday: 362 One week ago (7/14): 736 Newly reported cases T: 63K Y: 57K 7/14: 63K Newly reported tests T: 755K Y: 733K 7/14: 760K Positive test rate T: 8.3% Y: 7.8% 7/14: 8.3% — PolitiTweet.org

Nate Silver @NateSilver538

RT @Anna_Rothschild: Hello! @FiveThirtyEight is hiring a producer for Podcast-19! Looking for an audio storyteller with science/health jour… — PolitiTweet.org

Nate Silver @NateSilver538

RT @maggiekb1: Is COVID-19 making you exhausted, and you aren’t even infected? Me, too. Welcome to the summer of our discontent. https://t.… — PolitiTweet.org

Nate Silver @NateSilver538

I guess that was a dull tweet. But I do think people underrate the chance that Utah could become a swing state in another couple cycles, between Mormons not being super found of Trumpian politics and the non-Mormon voters there being pretty liberal. And look at these age splits: https://t.co/fmYmBXGYCL — PolitiTweet.org

Nate Silver @NateSilver538

I used to sometimes use a variable called "ldsevang" in our models where you could add together Mormon and evangelical voters as a proxy for religiously conservative states. But I don't know if that works anymore. Evangelicals shifted toward the GOP in 2016; Mormons against them. — PolitiTweet.org

Nate Silver @NateSilver538

@florian_krammer @WesPegden Agree with both of you. In my field at least, people can have a pretty good understanding of stuff that's fairly complex so long as you avoid confusing jargon. And they also have a pretty good sense for when they're not be leveled with. So no point in not being straightforward. — PolitiTweet.org

Nate Silver @NateSilver538

Hmm, I'm not sure how this plays out in practice. But in theory, if a vaccine is coming sooner rather than later, that ought to make the public *more* willing to undertake other interventions in the meantime, since they don't have an indefinite duration. https://t.co/PpqFSsEZCC https://t.co/glvqkcpPeb — PolitiTweet.org

Nate Silver @NateSilver538

Still, deaths rose sightly week over week, bringing the 7-day average to 754. We'll need a bit of luck to avoid going over 1K during Tuesday-Thursday, which typically feature much higher numbers than Sunday/Monday. — PolitiTweet.org

Nate Silver @NateSilver538

Mondays are typically slow, but I suppose this qualifies as less bad than most recent reports. One of the few days recently where there were fewer cases (57,948) than the same day a week ago (58,465). And the positive test rate was (barely) below 8% for the first time since 7/5. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 365 Yesterday: 523 One week ago (7/13): 327 Newly reported cases T: 58K Y: 64K 7/13: 58K Newly reported tests T: 735K Y: 769K 7/13: 722K Positive test rate T: 7.9% Y: 8.3% 7/13: 8.1% — PolitiTweet.org

Nate Silver @NateSilver538

A big part of why polling *averages* don't have a predictable bias (they miss in different directions in different years) is because different pollsters both overcompensate and undercompensate for their perceived mistakes in the last election. — PolitiTweet.org

Nate Silver @NateSilver538

It may be that some pollsters haven't fixed issues (e.g. education weighting) that led them to understate Trump's vote in 2016, but 2016 also created a market for some pretty spammy R-leaning polls with strong house effects. — PolitiTweet.org

Nate Silver @NateSilver538

Which one wins out? Well, we're actually trying to quantify that. But a decent prior is probably just that uncertainty should be somewhere around average. What I wouldn't want to do is take stuff from the 1st list that lowers uncertainty in my model *while ignoring the 2nd list*. — PolitiTweet.org

Nate Silver @NateSilver538

The case for high uncertainty: * Extremely high economic uncertainty * A huge amount of news, mostly around COVID, could create further shocks (yes, this can be quantified to some extent) * Uncertainties around voting mechanics during a pandemic may make polling (& voting) harder — PolitiTweet.org

Nate Silver @NateSilver538

Basically, this election presents a mix of reasons to think uncertainty could be higher *or lower* than usual. The case for low uncertainty: * Polls have been stable * High polarization likely = lower volatility * Relatively few undecided voters * No major third-party candidate — PolitiTweet.org

Nate Silver @NateSilver538

People might object to going back all the way to 1880. But in some ways, late 19c and early 20c elections are a *better* match for the chaotic conditions of an election amidst a global pandemic than say an election amid a standard postwar recession. — PolitiTweet.org

Nate Silver @NateSilver538

And if polls and "fundamentals" align (as they do this year; Trump is down by about the amount he "should" be) it might make you modestly more confident in the outcome, but not dramatically so. There is a long history of econ models being overconfident. https://t.co/uzhRzj9s2F — PolitiTweet.org

Nate Silver @NateSilver538

So while I'm a fan of using an economic prior in your presidential model (we do so in ours) it requires a fairly light touch. It might gently nudge the outcome from the current position in the polls, but it probably shouldn't do more than that. — PolitiTweet.org

Nate Silver @NateSilver538

And that assumes you know what economic conditions will look like in November, which in July you don't. This year in particular, we estimate that economic uncertainty is 2-3x higher than usual. — PolitiTweet.org

Nate Silver @NateSilver538

These variables explain about 30% of the variance election outcomes, which means there's 70% that they don't explain. That seems fairly realistic to me. And the average error is around 4 points of the incumbent party's vote *share*, which is around 8 points of vote *margin*. — PolitiTweet.org

Nate Silver @NateSilver538

I've posted versions of this before, but here's what you get if you run a model back to 1880 that tries to predict presidential elections based on a broad-based index of economic conditions and incumbency status, with an adjustment for polarization. It does *OK* but not *great*. https://t.co/46vzrSG129 — PolitiTweet.org

Nate Silver @NateSilver538

Basically, many things lately have reinforced a view of medium-term pessimism + long-term optimism and a lot of what I'm trying to figure out is when the medium term becomes the long term. Current prior is something like "in stages over the next 2 years"; won't be instantaneous. — PolitiTweet.org

Nate Silver @NateSilver538

(Also, I'm not sure if this counts as part of "the science", but the studies on immunity have generally had pretty optimistic results if you actually follow the research instead of relying on misleading headlines, though there are a lot of uncertainties/complications.) — PolitiTweet.org

Nate Silver @NateSilver538

The progress on the science front to fighting COVID continues to be at like the 95th percentile of reasonable expectations, even as the political/societal response has, uh, not been. https://t.co/pDHdpiD6Oz https://t.co/swqjmK2Abp — PolitiTweet.org

Nate Silver @NateSilver538

@briankoppelman If you properly account for lags, then the CFR is in the range of 1.5-1.9%. But still a huge gap between CFR and IFR because we likely aren't catching most cases (not enough tests, etc.) — PolitiTweet.org

Nate Silver @NateSilver538

0.3%-0.4% is still a very dangerous disease (wear a mask, please!). But, it's better than in the spring, where the weight of the evidence based on various estimates I was seeing was concentrated in the range of 0.7%-1.0% or thereabouts. — PolitiTweet.org