About the latest angst from 538...

[Buzz Clik]

fivethirtyeight.com came out with this late yesterday:

But those state polls? Not a lot of good news for Clinton. There’s more data showing a tied race in New Hampshire. And Clinton’s lead in Pennsylvania is down to about 3 percentage points in our forecast. Polls in Michigan have also been tightening, with an unusually large number of undecided voters. Polling in New Mexico has been tight enough that we’re now considering it a “state to watch,” although that may reflect an abundance of caution. Clinton’s numbers have held up better in Wisconsin and Virginia, while the data has been very mixed in Colorado.

http://fivethirtyeight.com/features/election-update-national-polls-show-clintons-lead-stabilizing-state-polls-not-so-much/

It's been interesting to watch 538's national trend as it hacks into into Hillary's lead every day:



But, we all know that the national poll is meaningless -- it's ONLY the state polls that matter. So, 538's article was disturbing.

As it happens, I decided to kill thirty minutes of my lunch hour yesterday, and I loaded all 538's "predicted margin of victory" data into a spreadsheet (current as of noon Friday EDT). I do this to let me push the numbers around and draw my own conclusions. So, being very familiar with their numbers, I found it quite curious to see all the hand wringing in this morning's article. So. I went back to the spreadsheet and entered all of today's numbers from 538 (as of 3 pm EDT). It was remarkable that, with only 6 exceptions, 538's predictions for every single state ticked toward Trump by at least 0.2% and as much as 2.5%, with an average of about 1%. This makes no sense. Polling data doesn't move like that in 24 hours with absolutely nothing else going on. This swing, by the way, was indeed ground shaking for Clinton. Her "solid" electoral votes (using this 538 data set) dropped by about 10; Florida and Nevada swung from virtual ties into the Trump column.

The only way this trend in 538's predictions could make sense would be if new polling data showed an across-the-board swing toward Trump. So, I checked the polls, and the only new data were from Survey Monkey, who released data for just about every state. There was movement both ways, but Florida and New Hampshire were most interesting. Florida, which 538 had in 24 hours swung from 0.2 for Clinton to 0.2 for Trump, showed no change in Survey Monkey, who continue to show HRC with a 2% lead. Only one poll in Florida during the last week has favored Trump (Remington), while the other seven polls favor Clinton. New Hampshire, according to 538, dropped from a 3% HRC lead to less than 2% overnight, whereas the only new poll (Survey Monkey) gives HRC a 10 point lead.

538 also brought up Michigan as an example of concern: they ticked their projection from 4.2% victory for Clinton to 3.8%. The new Survey Monkey poll was unchanged and a new PPP poll gives MI to HRC by 5%.

I am not buying into 538's pronouncement of a tightening race. It's stable. Clinton's lead is comfortable. If you take the straight polling numbers from pollster.com and add up the electoral votes, you get this:

Electoral votes in states with a lead significantly greater than margin of error:
HRC: 268, Trump 164

Add in those states at the margin of error:
HRC: 301, Trump 170

[StrictlyRockers]

1. I respect your analysis more than I do Nate Silver's at this point.

He's been so wrong lately. Sure he was accurate four years ago and eight years ago, but lately, nahhh.

[Buzz Clik]

4. I have no idea how he constructs his algorithms, and I am not commenting on them. HOWEVER:

...when his projections essentially all shift toward Trump with no data in the polls to support that shift, then I throw up my hands.

I have some very cynical explanations for this, but not now. I'm content to wait, relying more on these guys:

http://election.princeton.edu/electoral-college-map/
http://elections.huffingtonpost.com/pollster

[AgadorSparticus]

2. 538 doesn't smell right. Something just doesn't add up for me too.

Thanks for posting your findings!

[manicraven]

3. Great analysis! I gave up on 538.com. They're an outlier and not logically making sense.

[dogindia]

5. maybe old Nate has his finger on the scales

Talking to the FBI perhaps. This is strange and sad.

[redstateblues]

6. Nate is more interested these days in clicks to his site

More like the cable media.

[Buzz Clik]

8. ummmm...

.... yeah. That's pretty close to my cynical explanation.

[underpants]

7. Thank you for your work on this. I do have a question

Your totals at the bottom don't add up to 538.

[Buzz Clik]

9. True. I'll explain

A handful of states fall within the gray area of "statistical tie." I did not add their electoral votes to the totals.

But, for the sake of completeness:

All states with even a slight lead for a given candidate:
HRC: 322
Trump: 216

[underpants]

10. All right just checking. 246

Below is not the best version I have do this but, I think, it gives a good picture of the scope of the difficulty any Republican faces in a national election.


I'd heard numbers bandied about on where Democrats START national campaigns.
2012 is a good starting point. 2008 included the Indiana anomaly.

Obama 332
Romney 206
Take away from Obama's total
Nevada 6
Colorado 9
New Mexico 5
Iowa 6
Virginia 13
Ohio 18 AND Florida 29

You get 246. A Democrat has to pick up 23 Electoral College votes in any of those states to win.
The Republicans start with 206 and that includes N. Carolina.

[obamanut2012]

11. I agee with you 100% -- even though I didn't do as in-depth an analysis

as you, because I'm not a numbers/maths expert.

VERY well done btw!