 |
 |
The previous analysis consisted of almost entirely pre-RNC polls, and the Monte Carlo analysis, based only on polls, suggested that Donald Trump would have no chance of winning an election held then. We now have a half dozen new post-RNC polls, including polls in NC, OH, and NV. As you might expect, the post-RNC polls show Trump, now the G.O.P. nominee, doing better against the likely Democratic nominee, Hillary Clinton.
A change in this analysis is that I’ve tightened the “current poll” window from polls taken in the past three months to polls taken within the past month. This has two possible effects for each state. First, it makes the polling data more current, which is particularly important for an accurate portrayal of Trump’s performance. His image among Republicans has dramatically changed over the past 6 months, and we can expect he will start polling better now that he is the nominee. Additionally, the smaller “current poll” window will increase the uncertainty for many states just because there are fewer polled individuals included. Thus, we see some states turn from solid in the previous analysis to strong in this analysis.
Now, after 100,000 simulated elections, Clinton wins 97,074 times and Trump wins 2,926 times (including the 205 ties). Clinton received (on average) 314 to Trump’s 224 electoral votes. In an election held now, Clinton would have a 97.1% probability of winning and Trump would have a 2.9% probability of winning.
Update: Yes…Vermont is the wrong color on account of a data entry error for one VT poll. This has been corrected for the next analysis.
Did Trump get a convention bump? It appears so.
Here is the distribution of electoral votes [FAQ] from the simulations:
Some states stick out when compared to the previous analysis:
Arizona:
Trump dropped from 99% probability of winning AZ to 51% probability. What happened? First, we dropped from five current polls to three. But Clinton turns in a surprisingly strong performance in one of the polls. I suspect further polling will put Trump back into a strong lead.
Florida:
Here we have gone from 16 polls (over 3 months) to four polls taken in the past month. Clinton drops from 99.8% probability to a 56% probability of winning—essentially tied. This reflects that Clinton and Trump split the past four polls. The polling history over the past half year greatly favors Clinton, but it is not clear this will hold now that Trump is “real.”
Iowa:
With a modest reduction in number of polls, Iowa flips from Clinton (probability 71%) to Trump (probability 94%). In the five current polls, Trump leads in three (and by +17 in the most recent poll), Clinton leads in one, and the candidates are tied in one. Iowa has always been very close, but Clinton’s small advantage seems to have vanished.
Nevada:
Nevada is the other state that has flipped since the previous analysis, from Clinton with a 88% probability to Trump with 67% probability of winning. Polling is scarce enough that it is difficult to make too much of this either way.
North Carolina:
Recent polls have decreased from ten down to two, and Clinton has gained in her chances of winning from 76% to 91%. The polling history suggests Clinton’s lead is real.
Ohio:
Clinton has dropped in chances for this “classic” swing state from 83% to 64%. Clinton leads in three of the six current polls; Trump leads in two. The polling history has strongly favored Clinton, but the state might be tightening up.
Pennsylvania:
Another “classic” swing state that always goes to the Democrat. Clinton drops slightly, but much of this is the reduction in “current polls” from 10 to 3. Nevertheless, Trump does have a small (+2) lead in one of the three current polls. The larger trend suggests Pennsylvania is safe for Clinton.
The long term trends in this race can be seen from a series of elections simulated every seven days using polls from 26 Jul 2015 to 26 Jul 2016, and including polls from the preceding month (FAQ). Since the previous analysis, I’ve found about 65 older polls. These polls have little or no effect on the current analysis, but do sharpen the standings in the past.
An animated sequence of maps and electoral vote distributions can be seen here
Ten most probable electoral vote outcomes for Clinton (full distribution here):
- 309 electoral votes with a 2.37% probability
- 314 electoral votes with a 2.31% probability
- 313 electoral votes with a 2.10% probability
- 319 electoral votes with a 1.95% probability
- 331 electoral votes with a 1.81% probability
- 308 electoral votes with a 1.81% probability
- 320 electoral votes with a 1.80% probability
- 302 electoral votes with a 1.76% probability
- 315 electoral votes with a 1.63% probability
- 338 electoral votes with a 1.60% probability
After 100,000 simulations:
- Clinton wins 97.1%, Trump wins 2.9%.
- Average (SE) EC votes for Clinton: 314.1 (23.5)
- Average (SE) EC votes for Trump: 223.9 (23.5)
- Median (95% CI) EC votes for Clinton: 314 (267, 358)
- Median (95% CI) EC votes for Trump: 224 (180, 271)
Each column of this table shows the electoral vote total aggregated by different criteria for the probability of winning a state (Safe=100%, Strong=90%+, Leans=60%+, Weak=50%+):
| Threshold | Safe | + Strong | + Leans | + Weak |
|---|---|---|---|---|
| Safe Clinton | 103 | |||
| Strong Clinton | 140 | 243 | ||
| Leans Clinton | 59 | 59 | 302 | |
| Weak Clinton | 29 | 29 | 29 | 331 |
| Weak Trump | 12 | 12 | 12 | 207 |
| Leans Trump | 22 | 22 | 195 | |
| Strong Trump | 113 | 173 | ||
| Safe Trump | 60 |
This table summarizes results by state. Click on the poll count to see the individual polls included for the state.
| 1 | 0 | EC | # | Total | % | % | Clinton | Trump | |
|---|---|---|---|---|---|---|---|---|---|
| 2 | 8 | Votes | polls | Votes | Clinton | Trump | % wins | % wins | |
| AL | 9 | 1 | 3690 | 36.7 | 63.3 | 0.0 | 100.0 | ||
| AK | 3 | 1* | 435 | 37.5 | 62.5 | 0.0 | 100.0 | ||
| AZ | 11 | 3* | 1772 | 50.0 | 50.0 | 49.4 | 50.6 | ||
| AR | 6 | 1* | 623 | 43.3 | 56.7 | 1.2 | 98.8 | ||
| CA | 55 | 4* | 4034 | 63.9 | 36.1 | 100.0 | 0.0 | ||
| CO | 9 | 4 | 2083 | 54.0 | 46.0 | 99.3 | 0.7 | ||
| CT | 7 | 1* | 1024 | 53.2 | 46.8 | 92.8 | 7.2 | ||
| DE | 3 | 1* | 958 | 53.0 | 47.0 | 91.3 | 8.7 | ||
| DC | 3 | 1* | 1131 | 76.5 | 23.5 | 100.0 | 0.0 | ||
| FL | 29 | 4 | 4478 | 50.2 | 49.8 | 56.2 | 43.8 | ||
| GA | 16 | 1 | 450 | 48.9 | 51.1 | 35.9 | 64.1 | ||
| HI | 4 | 1* | 801 | 61.9 | 38.1 | 100.0 | 0.0 | ||
| ID | 4 | 1 | 402 | 34.3 | 65.7 | 0.0 | 100.0 | ||
| IL | 20 | 1 | 1023 | 59.7 | 40.3 | 100.0 | 0.0 | ||
| IN | 11 | 1* | 1779 | 44.9 | 55.1 | 0.1 | 99.9 | ||
| IA | 6 | 5 | 3286 | 48.1 | 51.9 | 6.2 | 93.8 | ||
| KS | 6 | 1 | 464 | 43.8 | 56.3 | 2.6 | 97.4 | ||
| KY | 8 | 1* | 1825 | 41.4 | 58.6 | 0.0 | 100.0 | ||
| LA | 8 | 1* | 1285 | 39.4 | 60.6 | 0.0 | 100.0 | ||
| ME | 2 | 1* | 366 | 54.6 | 45.4 | 88.8 | 11.2 | ||
| ME1 | 1 | 1* | 201 | 59.2 | 40.8 | 96.5 | 3.5 | ||
| ME2 | 1 | 1* | 162 | 49.4 | 50.6 | 45.7 | 54.3 | ||
| MD | 10 | 2* | 2657 | 65.2 | 34.8 | 100.0 | 0.0 | ||
| MA | 11 | 2* | 2086 | 62.1 | 37.9 | 100.0 | 0.0 | ||
| MI | 16 | 2 | 1476 | 52.6 | 47.4 | 91.7 | 8.3 | ||
| MN | 10 | 1* | 1139 | 56.1 | 43.9 | 99.8 | 0.2 | ||
| MS | 6 | 2* | 1783 | 42.3 | 57.7 | 0.0 | 100.0 | ||
| MO | 10 | 1 | 786 | 43.9 | 56.1 | 0.7 | 99.3 | ||
| MT | 3 | 1* | 1153 | 44.1 | 55.9 | 0.3 | 99.7 | ||
| NE | 2 | 1* | 1093 | 42.5 | 57.5 | 0.0 | 100.0 | ||
| NE1 | 1 | 0* | (0) | (100) | |||||
| NE2 | 1 | 0* | (0) | (100) | |||||
| NE3 | 1 | 0* | (0) | (100) | |||||
| NV | 6 | 2 | 958 | 49.0 | 51.0 | 32.7 | 67.3 | ||
| NH | 4 | 2 | 823 | 51.6 | 48.4 | 74.3 | 25.7 | ||
| NJ | 14 | 2* | 568 | 60.4 | 39.6 | 100.0 | 0.0 | ||
| NM | 5 | 1* | 774 | 51.8 | 48.2 | 76.6 | 23.4 | ||
| NY | 29 | 1 | 861 | 57.7 | 42.3 | 99.9 | 0.1 | ||
| NC | 15 | 2 | 1558 | 52.4 | 47.6 | 91.2 | 8.8 | ||
| ND | 3 | 1* | 1226 | 44.6 | 55.4 | 0.4 | 99.6 | ||
| OH | 18 | 6 | 4998 | 50.4 | 49.6 | 64.1 | 35.9 | ||
| OK | 7 | 1* | 873 | 41.5 | 58.5 | 0.0 | 100.0 | ||
| OR | 7 | 1* | 433 | 58.9 | 41.1 | 99.6 | 0.4 | ||
| PA | 20 | 3 | 3331 | 51.0 | 49.0 | 79.7 | 20.3 | ||
| RI | 4 | 1* | 886 | 57.0 | 43.0 | 99.9 | 0.1 | ||
| SC | 9 | 1* | 1380 | 44.2 | 55.8 | 0.1 | 99.9 | ||
| SD | 3 | 1* | 657 | 40.9 | 59.1 | 0.1 | 99.9 | ||
| TN | 11 | 1* | 2191 | 40.5 | 59.5 | 0.0 | 100.0 | ||
| TX | 38 | 1* | 852 | 45.1 | 54.9 | 2.2 | 97.8 | ||
| UT | 6 | 3* | 2090 | 47.5 | 52.5 | 5.3 | 94.7 | ||
| VT | 3 | 1 | 378 | 38.1 | 61.9 | 0.0 | 100.0 | ||
| VA | 13 | 3 | 1723 | 52.5 | 47.5 | 93.0 | 7.0 | ||
| WA | 12 | 1* | 584 | 57.0 | 43.0 | 99.2 | 0.8 | ||
| WV | 5 | 1* | 1187 | 33.4 | 66.6 | 0.0 | 100.0 | ||
| WI | 10 | 1 | 536 | 52.4 | 47.6 | 78.3 | 21.7 | ||
| WY | 3 | 1* | 690 | 29.6 | 70.4 | 0.0 | 100.0 |
* An older poll was used (i.e. no recent polls exist).
Details of the methods are given in the FAQ.
The most recent analysis in this match-up can be found from this page.
I’m curious as to why PPP polled NC so frequently. Were they doing polling due to the state issues in the news of late, and added the presidential polling to the state issue polling?
Thanks for doing this, Darryl.
538 currently has the race 53.2 to 46.8 in Clinton’s favor. Can you help us understand how to compare/contrast your results with Nate et al?
Zotz @ 2,
What I do is quite straightforward: I take current polls and simulate 100,000 elections in each state using only the polling data. This is closest to Silver’s “now cast”.
Silver goes beyond simply using the polls as raw data. He layers on many additional assumptions. He also weights polls according to their past performance. I don’t do any of that. That is, I only use actual poll responses for data; Silver uses other “stuff” as well.
As a consequence of his additional assumptions and/or poll weighting, Silver has a bizarre distribution of electoral votes. Check out his current “now cast”; it shows a distribution of electoral votes that goes from something close to ZERO electoral votes for Clinton to something close to 538 electoral votes for Clinton.
Put simply: That’s absurd. There is no way that either candidate is going to win all 538 electoral votes. Oklahoma, Idaho, and the like are NOT all going to go for Clinton under any circumstance. D.C., Vermont, Hawaii and the like are not going to all go for Trump under ANY circumstance.
Silver’s analyses have always had this problem that the distribution of electoral votes is way too overdispersed compared to anything that a straightforward statistical analysis would produce. I don’t know why.
That brings up another difference: I document my analyses completely, so that anyone with some programming skills can replicate what I’ve done. Silver’s methods seem to be proprietary, in that one could not replicate his analysis from publicly available information.
BTW: In the 2008 election cycle, my final analyses were slightly better than Silver’s. In 2012, Silver’s analyses were slightly better than mine. The reason? In 2012, the last poll in Florida was an outlier from a low-quality pollster. Silver “down weighted” the poll, I didn’t. So he had Florida slightly in favor of Obama (52% [or thereabouts] probability of Obama winning). I had Romney with a 63% probability of winning Florida. Silver painted the state light Blue and I painted it light Red. Florida was won by Obama.
Sloppy @ 1,
Probably because the company is based in Raleigh….
Trump most likely will not win in November. However, what are you progressives going to do if the current situation changes and Trump actually wins?
Politically Correct Commentary: I will NOT be voting for Trump, regardless of how things turn out between now and November.
Well done, Darryl, thank you for your great reply.
My mom (panicked 82 y/o Hillary supporter) hit me up with the secret sauce numbers this morning. This will help. Major good karma to you!
@5,
What will Republicans do if Trump wins, follows Putin’s orders and expands Medicaid, Medicare, and TANF, triples the debt, abandons NATO, guts the US Treasury, cancels all our trade agreements, and tanks the US economy?
Yeah, I guess Scalia II is worth all that. After all, Real ‘Murica’s gotta have them some “traditional marriage”.
#ConservaDerp
7
The Republicans will live with the choice, that’s what they’ll do.
What are you going to do, DR, if Trump actually wins?
Trumpty Dumbty says he will protect LGBT people from “foreign ideology”. This LGBT American is far more frightened by right wing ideology in this country and not just for myself. We know their tricks. Their spider and the fly tactics won’t work.
@8,
Precisely the same thing I’m doing today.
Hammering the Cheeto Jesus NO RAGRETS millstone around Republican necks.
#conservaderp
If Hillary wins in November, do you think she’ll have a celebratory cigar?
When Clinton wins in November they’ll be passing bags of Cheetos out…
at Gary Johnson’s triumphant “we surpassed 3% so spark it up” party.
@11
That’s why conservatives don’t have a decent comedy show on any network or even YouTube channel.
The joke was funny 20 years ago. Now? “is this thing on? We’re those tea partiers at the Early show?”
Got any new Ted Kennedy material? The fact that he’s getting close to a decade in the grave didn’t stop you before. You know what’s still funny? Raising a V with both hands and saying “I am not a crook?” “Wouldn’t be Prudent!” Hi-fuckin-Larious.
Gaffigan retired Hot Pockets. Galifinakis doesn’t do the Annie bit anymore.
Looks like Boob still has some work to do to push this election into the House of Representatives.
@5 @8 We will continue to live in this country, we will obey its laws, and we will be the best citizens we can be. And hope for better days ahead.
The mothers are speaking now, and it’s the most powerful thing I’ve ever heard come from a political stage.
@15
I can always be prouder. At some times more than others.
Czechsaaz @ 13,
I’m thinking that Sloppy has some good Teapot Dome material ready to roll…hoping he takes it to an Open Thread, though.
@18,
If we’re really lucky, maybe he’ll tell us all about The Petticoat Affair.
Not that the 3 EV’s matter, but Vermont is listed as 100% for Trump…..that can’t be right. I imagine the poll used had Trump, Clinton, AND Bernie on it?
Nate Silver’s analysis is more thorough! http://fivethirtyeight.com/fea.....dead-heat/ Trump at 285? We’ll see after the side splitting lying and fleecing speeches this week from the DUMMOCRETIN
People are upset at Fauxcohantas and The Bern for selling out!
JD @ 20,
Thanks for pointing that out. Indeed, I entered the Clinton and Trump numbers in reverse. It is corrected for the next analysis.
Puddy @ 21,
Not more thorough, just done using a different philosophy.
My analyses are done as a statistical purest would do them. I ONLY use state head-to-head poll data to predict an election held now. Polls represent data about a state’s preference, but with sampling error.
Silver uses many other types of information–national polls, correlations based on past elections, demographic data, etc.–and throws them into the mix to predict the election in November. He does provide a “now cast” that predicts an election held now–this is closest to what I am doing.
The other thing he does is “weight” polls based on the pollster’s past performance. I don’t do that. In fact, it is statistically dubious to do so because it mainly affects the dispersion of the outcomes.
If you look at Silver’s distribution of electoral college votes for his “now cast”, the possibilities range from almost 0 votes for Clinton to almost 538 votes for Clinton. Now look at my electoral college distribution (text version)
It suggests that, in an election held now, Clinton would have as low as 226 EVs (probability is 0.008%) or as many as 397 EVs (probability is 0.003%).
Whose distribution is “correct?” Given the polling data available today are there any data-based scenarios in which, for an election held now, Clinton would earn, say, only 3 electoral votes? How about Clinton getting all but 3 of the electoral votes? If so, what data are they?
Silver’s EV distribution suggests that there is a chance of either of these outcomes. Really? He predicts that there is a small probability that Clinton would take Alabama, Alaska, Idaho, Kentucky, Mississippi, Nebraska, Oklahoma, Tennessee, West Virginia, Wyoming AND almost every other state? Just based on actual data? I don’t think so.
My analysis, using only polling data, suggests that the results would fall from 226 to 397 votes for Clinton. It suggests she has no chance of taking Alabama, Alaska, Idaho, Kentucky, Mississippi, Nebraska, Oklahoma, Tennessee, West Virginia, and Wyoming.
I can “overlook” November predictions because there is a lot of uncertainty. But his “now cast” appears to be deeply flawed.
And that leads me to the final difference: I describe my methods in enough detail so that a reasonable programmer with a bit of knowledge about statistical distributions can replicate what I have done. Polling data are publicly available, and, in any case, I share my polling database. So anyone can replicate my findings. (This is a requirement in the scientific publication world that I believe should be adopted in statistical analyses of elections.)
Silver does not describe his methods in enough detail for someone to replicate them. (Or, I should say, I’ve never been able to find a complete description of his analytical methods.) In that sense his analyses appear to be proprietary.
During a Patient’s Substance Abuse Treatment Program. So, if you find yourself in issues like
Fentanyl addiction then without wasting a single second try to get
efficient addiction help. Prolonged use of Opioids
leads to dependence and addiction.