About to work legal proection.
This is it:
The national numbers look better than the state numbers, which say this is close.
A volatile election for sure:
>
The national numbers look better late. As I thought they would.
Pretty good agreement. I still think the third party breaks late for Clinton. But this looks like a 4 point race.
The cheat sheet. The big change is New Hampshire, which looks safer – though it is largely a function on one poll.
And Florida is still Florida. I think the EV numbers look good. One thing to note: a 2 point lead in Presidential elections is about 94% likely to hold. IN other words Trumps chances are very low – in my simulation he is at 6% – and I introduce a national error from polling back to 1968.
This is last. I could never square the red state performance by Clinton with the national polls. I still can’t/ Maybe its because red states break GOP, and as result the margin is overstated.
But dammit this worked in 2012. And it shows a much larger lead.
Your percentage of 94% for Hillary feels right to me. 538’s 70% is ridiculously low, as you’d expect as 538 has always had too high a chance for the underdog. But PEC’s 99% is too high – stuff happens in politics. Maybe you should start up an aggregation site.
Good job. A guy with a spreadsheet comes out with almost the same estimate of popular vote margin as a guy with a fancy model. The 538 model currently stands at 3.6% margin. There’s the difference between 71% probability and 94% probability of a win.
I keep hammering on the way the distribution looks in 538’s 10,000 trials. The lastest estimated value is around 300 electoral votes. But the model shows peaks at 330 and 360. There is a lot of variance in the data and actual random variable results can depart from the estimated value to any one of the states in the sample space. Those three values 300, 330, and 390 electoral votes are each 1% probability results for the model with the peaks between 270 and 360 having probabilities in the 0.5% range each.
This distribution of model results is really a bar graph, and the results are a discrete instead of a continuous distribution. Different rules apply than with normal distributions.
The probability of a win is not a confidence measurement. It doesn’t tell how good my data and calculations likely are relative to reality; it tells that there is enough uncertainty in the polling to allow Trump a 3-in-10 chance of winning the right combination of states based on comparison with the loaded data from a certain number of past elections. It is the brute force comparison of the current polling with the stand-in for “all past polling” that is the primary difference between Silver’s model and a spreadsheet analysis.
The expectation is that the 538 model allows more accurate standardizing of the data for a lot of explainable variation in the data by using the past data instead of rules of thumb or analyst memory or analysis to make those adjustments.
The result of the model runs right now say that the polling could be off but the expectation is a Clinton win in the neighborhood of 300 electoral votes and a 50-50 breakdown of the Senate. That likely means the House remains Republican.
But GOTV might change that, especially the Latino vote. And exit polls will only report on the people who talk to them. Those in fear of harassment likely are not going to be talking to pollsters. So we might or might not see GOTV efforts show up until the actual counts. And then the count protection effort starts in the midst of what is the most intense spin and gullible media in history.
He had the Cubs losing too.
Hey. thanks for all your diaries throughout this election season; always accessible and insightful. Here’s hoping.
Paul Krugman:
Glenn responds:
My response: Krugman says he’s not as much in touch with America as he thought. The majority are voting against and not for.
Much appreciation of your hard work this election.