1. News >
  2. Nate Silver >
3,128,356 followers   •   1,087 following   •   New York   •   http://fivethirtyeight.co
Editor-in-Chief, @FiveThirtyEight. Author, The Signal and the Noise (http://amzn.to/QdyFYV). Sports/politics/food geek.

Latest Scoops

1 is a good rule but there are many exceptions to 2
Procrastinating tweets are a negatively correlated with modeling progress at time N, but positively correlated at time N+1.
@ForecasterEnten Surprised that you left off Avenatti, Ossoff, Hillary Clinton, Ocasio-Cortez, O'Malley, and Howard Schultz, those are my top 6.
He's No. 2; Avenatti is No. 1.
What *would* be helpful for Trump is if those R voters are super enthusiastic and are highly likely to vote in November. But the evidence this year mostly cuts the other way and suggests that Dem voters are more fired up instead. Could be different in 2020.
Are either of these cases obviously better or worse for Trump? ¯\_(ツ)_/¯ As far as forecasting general elections go, they're probably about the same. They're *different* from one another, so it's worth discussing. But the overall number should still be the headline.
For instance, if Trump has 40% approval, it could be because:
—He's lost some support among R's and is merely treading water among independents
—He polls great among R's, but *really* bad with indies, and maybe some anti-Trump conservatives no longer call themselves Republicans
I don't like the trend of treating Trump's approval among GOP voters as the headline and his overall approval number as the subheadline.

Both are important, but the overall number is more important.

His approval among independents is also important, and rarely gets mentioned.https://t.co/9UXcrEduok
One of the essential strategies for triaging your news diet in the Trump Era is not to pay much attention to things that the White House is merely "considering". Some of them will happen but most won't, and they're often distractions from things that are actually *happening*.
2. Say you have a sample size of 5,000 cases, and of those cases, a dozen or so are missing or look as though they're miscoded. You might think, "That's not bad. I'll just throw those cases out and I'll be fine". And you probably will be fine if that's all there is to it.
Total(1) => 0.068015098571777 f_f_QM(2) => 0.060197830200195 indS(2) => 0.045740842819214 indM(2) => 0.012910842895508 indM_1(2) => 0.0030021667480469 indM_2(2) => 0.0007789134979248 indM_4(2) => 0.0018188953399658 indM_5(2) => 0.00331711769104 indM_6(2) => 0.0011749267578125 indM_7(2) => 0.0016539096832275 indM_8(2) => 0.0008091926574707 f_f_pTL(2) => 0.0062050819396973 f_f_dT(20) => 0.0059728622436523