Take a look at this graphic:




(courtesy @willjordan, YouGov, on Twitter)

There has been an expansion of one of the tails of the distribution toward Trump, slightly. The body of the distribution is more or less where it's been for a long time. And the sampling density of polls was much higher near the conventions. If we had more frequent polling I am fairly sure we'd fill in the middle of that distribution more!

People vacillate from JOY to TERROR here so quickly, far more quickly than the fundamentals underlying this election can possibly change.

Do yourself a favor, whatever emotion you are currently feeling about this election, examine it. Are you letting the potential consequences of HRC losing overwhelm the actual object probability she does lose? Then put that emotion away, it doesn't affect anything but you.

Throw away whatever your gut feeling is. Go to http://elections.huffingtonpost.com/pollster and count up the states where HRC has a lead. Add up their electoral votes. Try it a few times. Randomly drop a few states from her corner Use electoral-vote.com if you need to see a graphical representation. How many times does she lose the election? Divide this by the number of runs you try.

That's the only number you need to worry about.

Two models: one that only uses current polling, and one that takes the current poll as a starting point and and includes a random drift term...essentially projecting the current scenario forward to the election. See Original Model Writeup (below) for details.

Previous model writeups:

9/1/2016 update

8/16/2016 update

Update with drift included 8/6/2016

Original Model Writeup 8/3/2016

Model projections:

Probability of D win using current polling: 84.5%
average number of electoral votes: 312
median number of electoral votes: 314
most common electoral scenario: 317 electoral votes



Probability of D win projected forward to November election: 64.0%
average number of electoral votes: 288
median number of electoral votes: 289
most common electoral scenario: 289 electoral votes



Current Popular Vote Spread Estimate: +3.8% Clinton (estimated by multiplying statewide leads by turnout 2012 state turnout)


Analysis: Some minor tweaks to model, mostly in the code for grabbing data from HuffPost pollster. I missed an update last week, so I can't say any slide in HRC's poll numbers has been arrested. I can say that her numbers look about the same as two weeks ago, some slippage in the projected numbers, but that's due to changes in the polling variance over time. I use this variance to construct a daily random "drift" variable. In general, the variance in D leaning states' polling has gone up, the R leaning states show less variance. So it's more likely to randomly drift to a Trump win than a few weeks ago. The mean, median, and mode of electoral vote leads for Clinton haven't changed a lot though, which bodes well for continued stability.

Report on poll here.

I think this is indicative of the ?Live Caller" vs. Internet or "Robo-caller" divide again.
All we have had for weeks are crappy internet polls and robo polls. Now we have a traditional survey to examine.

Let's just hope that Live caller interviews are indeed better, as I suspect they are.

Perhaps late tonight too. The smarter pollsters wanted to avoid the labor day weekend as its effect on the sampling population being at home are unpredictable. So probably there were polls in the field Tuesday through Friday of this week. Maybe today too.

Anyhow it used to be a big deal to get a poll out before a Sunday talking head show, but I don't know how much they matter anymore. But definitely getting one out by Monday morning to help dominate the coverage seems likely.

I expect a consistent 3 point lead for Clinton in the average of polls taken over this period. Maybe 5 if we use a generous filter, but 3 is more likely given current trends.

Silver often claims the model is sensitive to small changes potentiall predictive of future trends.
Of course, I'm having a hard time seeing what happened in the data. Perhaps a large Trump sample rolled off?
It could be nothing. Give it two or three days to shake out.

Even if she is significantly ahead.

If it's a bit closer, but she is still ahead, this will happen more frequently. But it is a statistical certainty.

Examine the trend at http://elections.huffingtonpost.com/pollster/2016-general-election-trump-vs-clinton

There is a clear cyclical variation operating for the last few months in Trump's numbers. Also we appear to be at local maximum (his best performance is (usually) around 42%.

I would expect to see a split of 46H and 42T this week. Then I expect to see a slow decline in Trump's numbers.

On Edit: Adding image of poll trend

[img][/img]

Two models: one that only uses current polling, and one that takes the current poll as a starting point and and includes a random drift term...essentially projecting the current scenario forward to the election.

Previous model writeups:
8/16/2016 update

Update with drift included 8/6/2016

Original Model Writeup 8/3/2016

Model projections:

Probability of D win using current polling: 80.2%
average number of electoral votes: 300.3
median number of electoral votes: 300
most common electoral scenario: 300 electoral votes

Probability of D win projected forward to November election: 72.0%
average number of electoral votes: 285.6
median number of electoral votes: 300
most common electoral scenario: 312 electoral votes

Analysis: Some closing, but not in any range that really concerns me yet. Convention bump wearing off a bit.