Protected: Two ways to estimate primary outcomes without polls (transcript)
This content is password protected. To view it please enter your password below: Password:...
Post-Democratic convention bounce: 7 points for Clinton
So far, four five six pollsters have released national surveys using samples taken after the end of the Democratic National Convention, and have data from the post-RNC period. The median swing is a 7.0 ± 1.1 2.0 2.1 % (± estimated SEM) move toward Hillary Clinton.
Here are this year’s bounces, added to the previous graph.
Individual polls:
| Pollster | Pre-DNC | Post-DNC | Swing toward… |
|---|---|---|---|
| Raba | Clinton +5% | Clinton +15% | Clinton 10% |
| PPP* | Clinton +4% | Clinton +5% | Clinton 1% |
| Morning Consult | Trump +4% | Clinton +3% | Clinton 7% |
| CBS | Tied | Clinton +7% | Clinton 7% |
| CNN | Trump +3% | Clinton +9% | Clinton 12% |
| YouGov (4-way) | Clinton +2% | Clinton +5% | Clinton 3% |
The Pre-DNC measurement from PPP was taken in June, predating both conventions. The resulting swing therefore represents a combination of the Republican and Democratic conventions. However, since the median post-Republican convention swing was only 1 point, it was still informative and I included it. Since you can do medians in your head, you can see that this does not make a difference given the current data set.
YouGov leads its report with the 4-way (Clinton v. Trump v. Johnson v. Stein) result, showing a 3-point swing toward Clinton. The 2-way result shows a 2-point swing in the opposite direction, toward Trump. Again, the median is the same no matter which one of these is used. However, the error bar is affected slightly.
I have seen some sniffy comments in the news that Hillary Clinton’s post-convention bounce is smaller than Bill Clinton’s bounce in 1992. However, by current standards, Hillary Clinton’s bounce is large. As I wrote the other day, post-convention bounces have been small for presidential elections since 1996, which I suggest is a symptom of political polarization: voters get entrenched in their support.
Topics:
Is there a correlation between the relative post convention bump ( relative between the two parties , not time) and the winner in November? A logistic regression?
Please keep dragging me back to reality. Thanks, Sam.
I find it impossible that the same polling organization, RABA, is telling us the national race is 15 points for Clinton AND Trump is up 4 in Virginia. Does this not seem to be statistically impossible?
I was thinking the same thing. Neither of their results mesh with any other polling (or common intuition). When you are a new pollster and your polls are such wacky outliers in both directions, I think you should do some soul-searching.
Could be sampling error. You guys do some sqrt(Np(1-p)) math and see.
If you are saying that it seems impossible that Clinton is up 15% nationally while being down 4% in Virginia, I would agree with that. However, what you appear to be saying is that it is impossible for one pollster to obtain two outlying results that are in opposite directions from each other. I see no reason why such results would be impossible. The outlying results could simply be a factor of the particular samples used and not caused by any error that would consistently bias the results in one direction.
agree
The RABA poll was conducted via IVR July 26-27 ahead of both Kaine’s and Clinton’s Democratic National Convention speeches,
http://www.rabaresearch.com/documents/RABA-Virginia-Survey-Results-July-2016.pdf
I’d say its statistically very improbable rather than impossible.
You can really get into the weeds with cross tabs, but the VA sample doesn’t look bad. Random error strikes.
I hope this meets the data-driven threshold 🙂
after Dr Science’s comment i looked to see what had happened to the OFA GOTV machinery– it has morphed into Project Ivy.
It is supposed to migrate the benefits of OFA to downticket races.
here is a link–ivy.democrats.org/press.1.html
I also looked at jobs to see what they were hiring for–im sort of a data-nerd and i kno theres lots of recruitment efforts for datascientists and dataengineers right now. But Project Ivy is just looking for video professionals, junior designers etc– no open recs for datascientists. What that means to me is they have retained the talent that built OFA.
This may be a huge advantage going forward.
Thanks for the info, EWC! I’m guessing they also are looking to those pesky midterms, where the Dems *really* need to GOTV.
North Dakota just got their voter id law blocked– thats 3– plus NC and Wisconsin– dems problem is historically turnout.
we saw an example of datascience here with N’s google correlate analysis– Project Ivy has the potential to change the downticket races AND the midterms.
My question for Dr. Wang– can datascience beat the gerrymander?
According to Politico, Gallup saw a NEGATIVE Trump Bump.
“Last month’s Republican convention, however, is the only time respondents were overall less likely to vote for the candidate who was nominated. Previous lows for the Republican nominee were plus-2 with Mitt Romney in 2012 and plus-3 with President George W. Bush in 2004.”
http://www.politico.com/story/2016/08/poll-clinton-trump-dnc-2016-226507
Now that CNN’s in, the PPP poll does effect the median. Without it the median’s 8.5. And considering the last PPP was taken at a peak for Clinton, it also seems to be in line with a 8-9 point bounce.
Median of 1, 7, 7, 10 = 7, right? Median of 1, 7, 7, 10, 12 = 7, right?
@Michael
I’m saying that by including PPP it unfairly affects the average. With PPP the median is 7 as you said, but without it we have 7,7,10,12 = 8.5. And since the last PPP before this one was back in June right before the FBI announcement hurt Clinton’s numbers, a 1 point bounce from then is much, much better than a 1 point bounce from a week ago.
Median is not mean, Tony. To get the mean, you add up the values and divide by the number of values – that’s how you get 8.5. Median is the ‘middle value’. e.g the 3rd of 5.
The mean of 7, 7, 10, & 12 is 9—pretty sure Tony was just averaging the two middle values of an even number of values—the technique generally in use at PEC to establish a median in such cases.
Yeah, it was confusing because of the “now that CNN’s in” phrasing. CNN coming in didn’t change the median. Excluding PPP does.
Please remember to compile knife-edge congressional races for ActBlue donations, as you did in years past.
see left sidebar – went live today
Terrific! Thanks for that.
I was referring to your recommended “high leverage” or “high return” picks on races to donate to at ActBlue.
Left out: in the last election cycle you had an ActBlue page up as of the end of June:
http://election.princeton.edu/2014/06/28/high-leverage-races-for-2014/
Huffpost is dropping landline pollsters.
How will this affect PEC? It is getting the feed from Huffpo. On the face of it, landline polls are biased towards Republicans and dropping them would help Democrats.
But would it? How many such pollsters are there anyway, I wonder.
Do you think that having the conventions much earlier than they have been in the past will affect the polling? It’s a long way to the debates, which are usually the next big race-shaping events.
If Clinton is up 7 (post DNC polling median), why did the MM drop from 3.3 to 3.0?
Sam will probably correct me on this but my understanding is that the MM reflects state-level polls which tend to lag national polling trends?
yup
State polls are still kind of scarce, so Sam’s method smooths out a lot of rapid movement like convention bounces. Most of the wild gyrations in Nate Silver’s model lately are just the effect of national poll movement affecting his “trend line adjustment”, though people respond as if he’s got new state information.
The culprit for the MM movement appears to have been a single PPP poll that moved PA from Clinton +9% to Clinton +4%. Which just goes to show you that you shouldn’t sweat every little change in the MM either.
As electoral-vote.com pointed out the most interesting PPP poll result – at least for me – is this one:
“An astounding 74% of Trump voters think she should be in prison; only 12% disagree. Also, by a margin of 66% to 22%, Trump voters say that Clinton is a bigger threat to the United States than Russia. It is inconceivable that any of these voters will ever vote for her, no matter what gaffes Trump commits in the next 3 months. ”
Can Clinton convince enough undecided voters in the swing states?
This doesn’t mean anything. The definition of a “Trump voter” in this case is someone so far to the right that no logic or sane reasoning applies to them. The fact that the nominees are Clinton and Trump is irrelevant– those voters would never have voted for any Democrat.
Can Clinton convince enough undecided voters in the swing states?
That’s the only way to win any election. There’s a hard core of at least 40% to 45% of the electorate on either side who are simply unpersuadable. You don’t win elections by persuading hard-core republicans to vote for a Democrat, or vice versa, you win by motivating your supporters to get out in larger numbers than the other side’s, and by carrying more of the fraction of true independents in the middle.
Trump is basically counting on getting out his core supporters by being as uncompromisingly provocative as possible, while Clinton is relying on the same sophisticated GOTV operation that carried Obama to two victories. It will be informative to see which method is most effective.
alurin– that is the point of the Ivy Project– to extend the OFA GOTV machine from just the presidential candidate to downballot races.
If that works politics gets changed forever– im a bigtime data nerd– there are huge advances over the last four yrs in datascience and deep learning– u guys saw one technique here with N’s supercool use of google correlate!
it may work– dem problem is historically turnout– i think this is revolutionary!
i was kinda bored by this election cycle…but now im enthused.
@Josh
A lot of it depends on how many there are.
We should get an idea when the candidates file their formal monthly reports with the Federal Election Commission. That will tell us how many voters contributed dollars to his campaign, just too bad that one doesn’t know where they live:
http://www.nytimes.com/2016/08/04/us/politics/trump-fundraising.html
Do the pollsters have a mechanism to account for the overturn of voter suppression laws? That is, does it change their likely voter calculations?
nope
There are lots of factors that pollsters can’t account for (like, for instance, what happens when one campaign can’t or won’t effectively GOTV), and trying to pick and choose which ones you want to include is basically putting a thumb on the scale. Stick to the data that’s available.
As a long time junkie of this cite and some of its rivals I just have to comment on how sober Wang’s approach seems in comparison to some of its rivals. Very consistent since the win probablity opened. And this is just a part time gig for Sam, where others have made entire careers out of predicting elections. Then again, drama generates page views.
If you want to see drama go fivethirtyeight and check out their “now-cast” chart.
Drew Linzer tweets an interesting article at Vox.
https://twitter.com/DrewLinzer?ref_src=twsrc%5Etfw
Claims that post-convention bounces are the result of existing supporters being more willing to respond to pollsters rather than people actually changing preference from one candidate to the other.
That seems reasonable. I continue to believe that getting out your voters is far more important than changing a few minds.
Regarding the Raba poll. What does it mean that it is greyed out on the HuffPo page? It’s pretty clear that the Raba poll is not in the HuffPo’s branded chart. But, it is available if you do your own custom chart. I can tell that Raba is excluded because there is no poll in HuffPo’s model lower that 32% since March. If HuffPo is indeed excluding some polls from their model, even if they’re available for custom models, should we rely on them?
HuffPollster has started separately aggregating 2-way and 3-way (and 4-way?) polls. I think the graying out has to do with that.
Any historical perspective to assess impact of 1.2 % Q2 GDP growth on election?
Drew Linzer of Votamatic uses Q1 to Q2 GDP growth as a factor in his predictions, based on his fundamentals analysis of prior cycles. He says in a tweet that this year’s Q2 number “was a weak 1.2%. Not awful for Clinton, but not great either.”
Most (not all) of what we see as convention bounces are response bias – who picks up the phone. Online pollsters such as Yougov control for who responds to their surveys by partisanship in addition to age/gender/race. Using Yougov in a bounce calculation ignores what bounces are.
If you want to compare bounces across eras, comparing like for like with live telephone makes more sense. That’s why CNN and CBS showed some of the larger swings.
I can’t speak to the rest your questions, but I my guess is that the USC poll is excluded because it isn’t a random sample poll, but a panel. They interview the same group of people each time rather than trying to get fresh samples. This can be a great way to pick up movement, but has limitations when trying to determine the actual state of the race.
The USC-Dornslife-LA Times poll was off by about eight points in the California primary; they had Sanders up by one just before the election. The LA Times story on Sunday showed Trump with a seven point lead. Kevin Drum noted that the poll “had already been disappeared from the front page of the web edition, perhaps out of sheer embarrassment.”
Movement is precisely what we are looking at with the convention bounce, isn’t it? That’s why USC/LAT is of some interest here to me.
& Update: as of the Aug. 1 release just now (12:15am pacific), Trump has lost 5.1% since his peak in USC/LAT last Wednesday (released 12:15am Thurs) when Obama spoke. He’s losing about a point a day as it is a weekly rolling average.
Hard to tell what’s convention at this point, and what is ongoing Khan blundering. If we say it’s a ~7.0 bounce for Clinton in USC/LAT, that’s in line with the average of the others. If we say the bounce is more like 5%-6%, it’s in line with the others minus Raba. I would be inclined to exclude Raba, count USC ~5.5% and call it a 5.9% bounce.
OVERALL: On the eve of the GOP convention, Clinton was at +3.6% in both RCP’s and HuffPo’s average and is now +3.9% in RCP and +4.9% in HuffPo. (They seem to count CBS/YouGov differently, RCP includes USC and HuffPo doesn’t, HuffPo includes morning consult and RCP doesn’t. Neither include Raba, btw.) Overall, looks like a .3 – 1.3% Clinton advantage for conventions.
Movement may be what we are interested in here, but Pollster has other priorities. They want to capture the most accurate picture of public opinion possible. The USC/LA Times poll apparently doesn’t use a methodology they approve of, so they exclude it.
They have also begun excluding polls that use landlines only for the same reason; they exclude too many people, so they do not give a sufficiently accurate picture of opinions.
How comparable is the USC/LAT polling to the RAND poll of previous election cycles?
I think that bounce is still developing.
In perhaps the most shocking poll result I have seen to date:
Clinton – 36
Trump – 35
UTAH
(Hinckley Institute-Salt Lake Tribune poll)
I have heard talking heads mention Utah could be surprisingly close, but I still didn’t believe it. I kind of still don’t. Strange year for elections, that’s for sure.
It would be interesting to see multi-way polls in Utah. Mormons hate Trump, but are they really going to pull the lever for Clinton? Might be an opportunity for an astute 3rd party…
Oh, that’s right, we don’t have any astute 3rd parties.
The Washington Post has been unable to find this poll. I’m not sure it exists.
I heard that Mia Love’s internal polling showed Clinton, Trump and Johnson in a 3 way tie in her Salt Lake City District.
Allurin, based upon this poll, it appears that Johnson or Stein, most likely Johnson, could win votes in Utah with Clinton leading at only 36%.
I think this was corrected to a 35-35 tie, with Johnson pulling in something like 13% of the votes. That still leaves a large number of undecideds.
That poll showing Clinton leading in Utah? It doesn’t exist:
In an email to The Washington Post, the [Salt Lake] Tribune‘s Dan Harrie confirmed that there in fact wasn’t a new survey from the paper.
Ha. I got suckered by faulty news!
Is there a national popular vote victory margin that can be said to essentially guarantee flipping either or both houses of congress?
Based on how strong down ballot choices correlate with presidential picks, if Hillary wins the national popular vote by more than 5-6% it would seem very possible, if not likely, that Congress would flip.
Sabato’s Crystal Ball only considers 54 of the 435 House races to be worth rating. The other 381 are non-competitive, immune to almost any magnitude of landslide at the Presidential level. Democrats would be five seats short of a majority even if they win every race that is currently rated in their favor, as a toss-up, or leaning Republican. It seems unlikely that even a Johnson-Goldwater result would flip the House with current districting.
Not really true. Those ratings change with time, and some probably will by October.
I wonder if it wouldn’t be better to adopt a pre-RNC baseline. I just redid Sam’s table using July 1-15 as the baseline. In my back-of-the-envelope calculation Clinton gained two points (this is the median out of seven surveys with current and pre-RNC data). Regardless of the baseline, I think this is a good bump for Clinton.
Is the DNC giving Hillary a favorable bounce?
“Trump gets two-thirds of Republicans in the Granite State and only slightly more in Pennsylvania — and this is in a head-to-head match-up. More remarkably, Clinton’s net favorability in each state — the percentage of those saying they view her positively minus those who view her negatively — is about even. Trump’s is minus-30.”
https://www.washingtonpost.com/news/the-fix/wp/2016/08/04/state-polls-mirror-the-national-trend-trumps-campaign-is-wobbling/