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.

https://today.yougov.com/news/2016/08/30/clinton-leads-5-voters-label-trump-inconsistent-im/

Clinton leads Trump 42% to 37% in a four-way contest that includes Libertarian Gary Johnson and Green Party candidate Jill Stein. The gap between the Democrat and the Republican remains five points, 47% to 42%, when those who favor other candidates, or aren’t sure what they plan to do, are asked their preference between the two.

Poll actually shows a larger Clinton lead than last week.

Statistics, people. Some days you get a slew of results you don't like. For no real reason at all.

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:

Update with drift included 8/6/2016

Original Model Writeup 8/3/2016

Model projections:

Probability of D win using current polling: 93.8%
average number of electoral votes: 326.4
median number of electoral votes: 330
most common electoral scenario: 337 electoral votes

Probability of D win projected forward to November election: 79.3%
average number of electoral votes: 299.3
median number of electoral votes: 301
most common electoral scenario: 293 electoral votes

model appears to be converging to consensus estimates.

For the original model writeup, see here.

I made some tweaks to the model, and allowed for a random "drift" in the current numbers, based on the average variance per day (drift variance) in polling data from each individual state. This allows the model to calculate a "forward likelihood" of an outcome assuming the values drift from the current electoral estimate in a stochastic fashion. Drift variance in polling numbers are estimated for each state directly from polling data from the beginning of 2016. For states with no polling data, we average over all known state drift variances and use that value. So we will get two win outcomes: an outcome if the election was held today...and the probability of that same outcome happening in November.

With this in mind the model generates the following outcomes:

If the election were held today, the probability of a Democratic win is
87.3%

and the probability of an R win is
12.7%
with an average of 308 electoral votes, and a median of 310 for the Democrats. The most likely outcome for Dems is 322 EVs


####################################################


The probability of a D win in November is:
73.6%

and for the Rs:
26.4% (of course)

with an average of 289 electoral votes, and a median of 291 for the Democrats. The most likely outcome for Dems is 296 EVs.

So everyone and their brother has an election model these days I figured I should try and put one together too.

It's a fairly simple model and it samples polls via the Huffington Post Pollster API. I use a median filter moving average to estimate the likely spread between Democratic and Republican (D-R) percentages for each state for which polling is available.

There are many states for which no polling is available yet. Default D-R spreads for each unknown state are assigned using a heuristic: Deep South states have a strong (20 point) bias in favor of Republicans...Unpolled New England states have a 5 point bias toward Democrats. Each unpolled state has an implicit variance in this estimate derived from the average margin of error in polled states.

For states with polling, state variance is estimated from the average variance of the most recent polls from that state (usually between 3 and 5 polls). If there are no polls in the current monthly interval, we estimate a trend from older data and calculate the likely D-R spread based on this trendline.

Electoral votes are assigned to Democrats for positive D-R speads, Republicans for negative spreads.

The model is run 10000 times, with random noise added to each spreads based on the variances estimated above.

Today's model run predicts an 80.6% probability of a D win and 19.4% probability of an R win, with an average of 290 electoral votes for the Democrat and a median and mode of 292 electoral votes.

An histogram of outcomes for Democratic Electoral Votes - Republican Electoral Votes is shown below. Outcomes above zero are scenarios where Democrats win the electoral vote.