Swing state polls just in (all from Greenberg Quinlan Rosner, whoever that is):
OH Trump +2
PA Clinton +8
Nevada TIE
NC Clinton +4
Overall pretty good for Clinton compared to recent polls for those states. She’s still strongly ahead on enough states for 270, with the closest being NH and CO.
Yes I am clearly disputing that.
The data I am looking at is the HuffPo graph I previously linked to. [This one.](http://elections.huffingtonpost.com/pollster/2016-general-election-trump-vs-clinton/edit#!minpct=0&maxpct=100&mindate=2014-01-01&maxdate=2016-09-20&smoothing=moderate&showpoints=yes&showsplines=yes&hiddenpollsters=&hiddensubpops=A,A - D,A - i,RV,RV - D,RV - i,LV,LV - D,LV - i&partisanship=S,P,N&parties=D,R,I,N&selected=clinton,trump&fudge=1&except_questions_with=johnson)
What it shows is that there had been a slight gradual rise in GOP choice of Trump from April through into July, from averaging 76ish to 79ish before the GOP convention and a bump with the convention to nearly 81%. Then drop to 77 to 78 over most of August coming back up to 81ish by early September (over 80 on 9/1) and not much movement from there since. Today’s number on that curve 81.2%.
Apples to apples, keep with HuffPo only? Okay. A fair comment.
8/5 Clinton +8.3. Closest on HuffPo 9/12 Clinton +4.0. Sure other aggregators called it closer but we will stay true to the exercise. A 4.3 point swing.
As per the above link the shift in GOP specific support during that time was 78.2 to 80.9. A 2.7 point shift within less than a third of those polled in most polls listed - which would move the total result by 0.9% - roughly about a fifth of the total move.
Again, the thesis fails.
Maybe it is more Independent identifying GOP-leaners returning from undecided or other to Trump? But then how to explain the previously documented point that the total number of those not choosing either Clinton or Trump has stayed pretty rock solid?
OTOH the mean between the HuffPo peak and the bottom 8/5 to 9/12 is about Clinton +4.5, pretty near the long term mean of the race. Ranging +/-3 ish from that mean with occasional short-lived breaks beyond 3, driven by news cycles shifting the loosely attached and unattached in and out of groups is completely consistent with what we have seen.
Which poll has Clinton +4 in NC – that would be news to me. Not saying there isn’t such a poll but the ones I’ve Googled in recent days suggest a virtual deadlock.
ETA: Okay, just noticed the source. Not sure that’s really a trustworthy poll.
Guess I just looked at 15th to 15th. Still, if you look at the actual listed polls that average looks messed up by the big Monmouth +13 outlier poll. No other poll put her above +7/8. Most a fair bit lower.
538 gives them a B-. Not great, but not terrible – lots of posters get worse than that by 538.
Who cares?
You sure you are looking at RCP averages?
In the 4-way there was the +14 point Marist just a few days earlier and ABC’s +8. And in the 2-way Marist had been +15 and there were several +8 and +9. Heck using the 2-way take the last 8 polls, throw out the highest and the lowest as outliers and average the rest and you still get that she was +7.7. +7.9 was not a result of a single outlier poll.
That said I made misstatement in my post. Clinton +4.5 is the middle for RCP top to bottom in that time period, not for HuffPo’s technique on their standard presentation. HuffPo, which saw Clinton bottoming out at +4.0, has a mean for that time more like 6.1%
Actually, I don’t know how you are looking at past averages. Maybe it’s because I’m on my phone but I can’t click on the chart or anything. But you can’t be denying that the Monmouth is an outlier. Around August 8tb:
NBC News/SM 8/8 - 8/14 Clinton +6
Rasmussen Reports 8/9 - 8/10 Clinton +3
Gravis 8/9 - 8/9 Clinton +5
Bloomberg 8/5 - 8/8 Clinton +4
Reuters/Ipsos 8/6 - 8/10 Clinton +5
Economist/YouGov 8/6 - 8/9 Clinton +6
Monmouth 8/4 - 8/7 Clinton +13
NBC News/SM 8/1 - 8/7 Clinton +6
ABC News/Wash Post 8/1 - 8/4 Clinton +8
McClatchy/Marist 8/1 - 8/3 Clinton +14
NBC News/Wall St. Jrnl 7/31 - 8/3 Clinton +9
You are saying that the Monmouth and the Marist one don’t strike you as unusual?
Sorry about that everyone. Apparently, sleep deprivation, 538, and a 1.5L bottle of Shiraz don’t compliment each other as well as I thought they would.:smack:
You’re still not measuring the time period I mentioned, and you’re still assuming that none of that shift was Clinton to Trump. If you measure post-convention bump to mid-September, and you assume some small percentage of the shift was Clinton to Trump, you can explain a majority of the movement with the GOP shift alone.
The bigger picture point is this: There’s not very much evidence that there are millions of voters out there whose opinions shift week to week. Much more plausibly, most of that movement is a combination of (1) noise–both statistical and because of differing pollster methodologies; (2) differential response rates based on enthusiasm; and (3) differential movement of the fundamental tectonic plates of modern presidential opinions–shoring up partisan support and third-party and undecided vote slowly splitting out to the candidates.
Are there any political scientists or retired political professionals (who therefore no longer have a financial incentive to say otherwise) who think winning the weekly news cycle matters, or that a story that lives and dies in a single week actually moves voters? That would be worthy qualitative evidence to weigh, if it exists.
Carnal K,
I am saying that the RCP rolling average on 8/9 of +7.9 was not driven by one (or two) “outliers.”
Look. Here’s all the two-way polls done containing dates within 8/1 through 8/9, no dates before and none after:
Economist/YouGov 8/6 - 8/9 911 RV 4.2 48 41 Clinton +7
Bloomberg 8/5 - 8/8 749 LV 3.6 50 44 Clinton +6
ABC News/Wash Post 8/1 - 8/4 815 RV 4.0 50 42 Clinton +8
McClatchy/Marist 8/1 - 8/3 983 RV 3.1 48 33 Clinton +15
NBC News/SM 8/1 - 8/7 11480 RV 1.2 51 41 Clinton +10
Average is 9.2. Throw out the top and the bottom as outliers and the average is 8.3. Median is 8. Calling the rolling average as 7.9 is not a fluke of one outlier.
Your combined 4-way and 2-way list as above? Average is 7.3. Throw out the top and the bottom numbers as outliers and average? Is 7.0. Still within 1 point of the 7.9. Okay, use the median of the combined list and the median drops to 6.
RP I have never heard a single expert ever claim that news cycles and events will have no impact on polls (which seems to be what you are arguing now).
The issue is who the news cycle will have an impact on and how many there are who can be impacted by it versus those who are firmly decided. And how long those news cycle peturbations last.
My belief, oft repeated by now, is that the electorate is pretty firmly polarized and entrenched with a relatively few who are only loosely attached to one of the two main candidates. I see that when those loosely attached are distributed without major news cycles impacts the race settles to roughly Clinton +4 (I did think +5 but have been corrected). There are enough of those loosely attached that news cycles can move enough of those people such that poll averages can vary plus or minus 3 points from that range, but moving beyond that range is a bigger lift that would require a more seismic shift. Thus polls ranging between as high as Clinton +7 to as low as Clinton +1 are going to happen but excursions outside of those ranges will be short-lived.
Turnout based on enthusiasm and on organization is not captured in that though. Who actually is likely to vote is the wild card in my mind.
You’re stating a more extreme version of my thesis, but I take your comments to mean you’ve never seen the political science literature on this. My sense is that the weight of opinion is on the skeptical side about whether news events that come and go before Election Day matter very much. My sense is that most think it’s like the conventions: a bounce that matters for fundraising and some other things but that inevitably fades. I’ll see if I can dig up some articles for you. Here’s one to start: http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf
And now having said that, I think I have conflated two arguments.
One argument is that the polls themselves have moved week-to-week mostly from fundamental forces, conceding that news cycles account for some level of volatility in a less smoothed trendline.
The other argument is that the short-term volatility reflecting news cycles isn’t actually voters changing their minds about whether to vote or who to vote for.
I believe both things, but maybe we are just debating the first?
Yes.
In fact I have been consistently arguing in multiple threads the second position. News cycles move polls up and down some but this race is, like most other races in our increasingly polarized electorate, fairly stable. The cumulative effects of the news cycles can have some net impact on some number of loosely attached and undecided voters and on turnout which will matter in a close election but my read is that it operates above and below a not extremely wide range. Again, it can be wide enough depending on the election. That’s the point of anchoring to a Bayesian prior which returning to the point of this thread is “the fundamentals” that Silver has used that anchor to a close race, or the long term polling trend that Wang uses which anchors to Clinton +4 to 5. It is also what methods like PollyVote attempt to correct for by combining multiple methods above and beyond polling and fundamentals alone - a hope to reduce the noise and somewhat fictitious (but “newsworthy”) volatility without losing true signal.
The article you linked to concluded that the “temptation to over-interpret bumps in election polls can be difficult to resist” and I completely concur. In this case what we are debating is whether or not you are over-interpretting the drop from a post-convention bump to a (transient) news cycle driven pothole as reflecting actual long term fundamental forces, specifically identifying GOP partisan return to the barricades as that fundamental force.
Another question is at what point we start to discard the Bayesian prior, whatever we believe it is, and believe that aggregated recent polls alone give us the information we need.
I am a little bit skeptical of this one – not that it couldn’t happen but IIRC a poll released last week showed Clinton up by 8 in PA and there’s really nothing in PA or nationally that would suggest a sudden Trump surge over the past week.
But even if you’re optimistic, Clinton is probably, realistically, only up by 3-4 points in the Keystone State. As I’ve been saying, there really, truly is a LOT riding on tomorrow’s debate. If there’s a virtual stalemate, we might see a slight shift either way, but the situation would be recoverable for either candidate. However, if there’s a clear winner and loser like there was in the first Romney-Obama debate, then we may be well on our way to knowing the eventual winner.
The point is not being skeptical but an informed consumer who understands how to place a single poll that reports a 5% MOE in context.
Overall polling there is running Clinton +4 to 5 (closer to 5). That’s not optimistic it just is what the average is. This house with an instrument that has MOE +/-5 has been above that average once by 3 to 4 and below that average once by 2 to 3. It’s really what you’d expect if the number was a completely stable +4 to 5.
Now that said, if you want to worry, worry that we have so little recent polling that we really don’t know. PA and OH being very divergent is odd after all. Ohio is more densely polled and it is not impossible that PA has in fact been less strong than the polls have looked. If so then I will join in the wringing of hands.
Yeah I think Clinton +4 or 4.5 seems about right, just based on the polling over the past 14 days (or what I can recall of it). I just hope it doesn’t reflect the start of a downward trend. I don’t honestly see why it would because nothing in PA that I know of should have moved the dial.
Unfortunately, I am less skeptical of this poll than the one in PA.
One thing that jumped out at me was that Trump rates a lot higher when asked about how “exciting” the candidate is. What struck me is this: Trump is getting loads of bad press, but Clinton has been relatively quiet, and I’m not sure that this is a good thing. But what do you do? How do you make noise? She’s back on the trail, so it’s not like she’s not saying anything. It’s just that all we hear is Trump. All you need to do is go to Google News and he’s almost always at the top of trending news topics – almost every single day. I mean shit, is that how we’re going to decide the presidency – by who’s trending on Google and who excites us more?
For all practical purposes, the race is nearly dead even. Tomorrow’s big.
For practical purposes, Clinton has held the lead throughout. The just released WSJ/NBC poll includes how enthusiastic supporters are and the candidates were close to even so I dont think Trump’s numbers on election day will be above his polling.
Clinton did hold a lead once – she holds it no more. They are tied.
If Donald Trump wins the debate tomorrow, he’s probably our next president.