States with new poll data added since the last update: Wisconsin, New York, North Carolina, Pennsylvania, Ohio, Michigan, Virginia, West Virginia, Arizona, New Hampshire, Missouri, Kansas
With the logic change out of the way, the race moves on. With the latest batch of polls, one state changes category, and two more contribute to a change in the tipping point.
Electoral College Trend
Despite a few bad polls for Trump right after the convention, the general trend in Michigan has been a tightening race since the beginning of July. With the latest, Clinton’s lead falls to only 4.4%. For the first time since March, Michigan is looking possible for Trump, so we now include it in his best case.
If Trump wins all the states he is ahead in, plus all the states where Clinton leads by less than 5%, he now wins by 56 electoral votes.
(Note the transient spike on the chart as well. This was caused by one poll added in this batch briefly bringing Pennsylvania back into reach for Trump, but another poll added in this same batch increased Clinton’s lead again right away, so there was no net change due to Pennsylvania this time.)
Tipping Point
Ohio and New Hampshire were and still are “Weak Clinton” states, but they wiggled around a bit with the latest polls and moved the tipping point from Clinton by 4.4% in Ohio to Clinton by 3.8% in Ohio, an 0.6% move toward Trump.
Overall
In both the Trump best case and the tipping point, we see Trump clearly reaching a peak during the conventions, then plummeting the first half of August, then recovering ever since. He isn’t quite back to where he was before the conventions, but he is getting close.
So far Trump is making a lot of states that were not close a few weeks ago close. But no states have crossed the center line in the last couple of weeks. We have seen this before. There are a variety of blue states where Trump seems to be able to come close, but he is having a hard time pulling them across the line into the red zone.
Since the conventions the “expected” case has been in the range between Clinton winning by 144 and Clinton winning by 188. Trump hasn’t done better than losing by 144 electoral votes since before he was locked up the Republican nomination. The test for any continued Trump rebound will be if he can start to move not just his best case, but the expected case as well. Can he actually flip states? Or just make them close?
The electoral college margin in the expected case is deceptive, because these numbers can change very quickly. The tipping point is the thing to watch at the moment. At a 3.8% Clinton margin, less than 1 out of 50 people have to change their minds to flip the outcome. The public is polarized, but 1 out of 50 isn’t all that much in the grand scheme of things. You can imagine events that would flip that many people if Clinton has a bad week.
Historically though, Trump is very near his previous ceilings. Every other time he has reached these levels, he has fallen back down fairly quickly. Will this time be different?
67.2 days until the polls start to close.
Note: This post is an update based on the data on ElectionGraphs.com. Election Graphs tracks both a poll based estimate of the Electoral College and a numbers based look at the Delegate Races. All of the charts and graphs seen in this post are from that site. Additional graphs, charts and raw data can be found there. Follow @ElectionGraphs on Twitter or like Election Graphs on Facebook to see announcements of updates or to join the conversation. For those interested in individual general election poll updates, follow @ElecCollPolls on Twitter for all the polls as they are added. If you find the information in these posts interesting or useful, please consider visiting the tip jar.
[Blog Post] Electoral College: Trump Still Rebounding https://t.co/YxTE8C7AuM
RT @ElectionGraphs: [Blog Post] Electoral College: Trump Still Rebounding https://t.co/YxTE8C7AuM
Michael Loomis liked this on Facebook.
You’re point about “1 out of 50” voters switching to Trump flipping the election is a very salient point and one that I think not many people appreciate. That’s one of the reasons I like that the “bubble” is always included in the electoral college results. Correct me if I’m wrong, but I read the bubble as being a tipping point threshold: if 1 out of 40 people switch (2.5%), this would be the outcome.
Have you considered making a single graph that merges the electoral trend line and the tipping point graph? Currently the bubble is just a single tipping point threshold. The electoral trend line could be made fuzzier by showing successively lighter bubbles around it, representing the outcome given an x% change.
You wouldn’t need to calculate every possibility from 0.0% on up because the tipping point states are discrete, not continuous. Instead, you’d simply have gradation lines: 0.8% this way is Iowa, 0.8% the other way, Georgia. However, since they have a different number of electoral votes, you’d be further up or down from the trend line.
[Yeesh, I shouldn’t post from my phone. Just noticed that Swype switched my very first word. This must be karma for feeling annoyed at other people using “you’re” for “your”. Note to self: be kinder about grammar.]
The tipping point line really doesn’t correspond directly to anything on the Electoral College trend graph. It measures something different. It is true that on the electoral college trend graph the “best cases” are the boundaries of where states are at 5% either way, and the center line is where states pass the zero line. It would be possible to draw other lines at 4%, 3%, 2%, and 1%, but the graph would just get really cluttered and hard to read. Similarly if I tried to show actual states on that line. It is a lot clearer just saying “under 5% could go either way, so I show the full range of possibilities where the states under 5% go either way”.
I’ll note an additional thing though. It is 1 of 50 overall voters changing their mind. But of course to flip the election from Clinton to Trump, all of those voters would have to be Clinton voters. The percentage of Clinton voters that would have to swap is of course bigger. (Actually, it could be undecideds too. I sometimes put in that disclaimer. It is X% of voters have to change their minds OR the undecideds could break for the currently losing candidate.)
The tipping point line really doesn’t correspond directly to anything on the Electoral College trend graph. It measures something different. It is true that on the electoral college trend graph the “best cases” are the boundaries of where states are at 5% either way, and the center line is where states pass the zero line. It would be possible to draw other lines at 4%, 3%, 2%, and 1%, but the graph would just get really cluttered and hard to read. Similarly if I tried to show actual states on that line. It is a lot clearer just saying “under 5% could go either way, so I show the full range of possibilities where the states under 5% go either way”.
I’ll note an additional thing though. It is 1 of 50 overall voters changing their mind. But of course to flip the election from Clinton to Trump, all of those voters would have to be Clinton voters. The percentage of Clinton voters that would have to swap is of course bigger. (Actually, it could be undecideds too. I sometimes put in that disclaimer. It is X% of voters have to change their minds OR the undecideds could break for the currently losing candidate.)
Dangerous comment about 1 in 50 I think. It’s a bit ambiguous (is that the word?)
The problem is that 1 in 5 could switch to Trump and not change anything in the election. (Maybe a slight exaggeration but the point stands).
Why? The electoral college…
Every non-Trump voter in TX, WV, ND, SD, AK and so on could switch and give him 100% in all ‘red’ states. Trump would win the popular vote and have gained numerous voters.
Unfortunately he’d still lose, just as Gore did in 2000.
I tend to think that the current tightening of polls is mostly just down to convention bounces finally wearing off. Probably not all but most.
I believe its around now that pollsters switch from generally using RV to using LV. So I think some hiccups are likely over the coming weeks as this happens. Some states might start to look well out of reach and others may tighten substantially.
I wouldn’t be surprised if some strange states begin to look close and flippable for both sides, say TX or WI for example. After all, stranger things have happened
Yes, 1 in 50 is not a well qualified statement. We’d also have to add in that it is assumed to be a homogeneous change over the entire nation (Nate Silver has good evidence that this is a reasonable assumption). And, to be perfectly pedantic, we should be clear that we’re talking about the NET change: it’s easy to imagine 1 in 50 people changing their minds on anything, but we have to subtract the changes going the other way.
I’m a bit unclear on the undecideds swinging for Trump. If there’s a 3.8 percentage point (pp) difference, and Trump gets a 2pp boost, he only wins if Clinton also loses 2pp. So, 1 in 50 has to be talking about voters who flipped from Clinton, right?
My idea for merging the electoral college and tipping point graphs is not to clutter the graph with lots of lines, but to make the trend line appear fuzzier. You can think of it as shading in the bubble with darker colors for the more probable outcomes, with the darkest being the actual trendline (0% change). (Well, “probable” isn’t the right word, but whatever the metric is that the tipping point measures.)
I think this would actually make a beautiful, uncluttered graph that would also be easy to understand. But, perhaps I’m missing something.
You are right, the 1 in 50 is not well qualified. In previous posts where I have made similar statements, I’ve actually said a lot of the “if there was a uniform swing and you don’t consider undecideds, etc, etc”. I didn’t this time, and maybe should have. I probably won’t use that exact formulation again. Just too many qualifiers to explain exactly what I am really talking about. The main point is simply that a 4% margin is actually relatively small. A lot of things can erase a 4% margin. This is why I consider anything less than 5% to be a swing state that could go either way. Not necessarily that I don’t believe the number, but because any number of events could erase a margin that small in a very short period of time.
As for the undecideds, one thing that focusing on the margin does is effectively assume the undecideds will break in a way that doesn’t change the picture much. Looking at just the margin, you can’t distinguish the case where you have a 5% margin, but only 1% undecided from the case where you have a 5% margin, but 30% are still undecided! The second cases is actually much more volatile than the first, but a margin only view (like this site) doesn’t give you that insight. As long as the undecideds end up breaking pretty much like the existing distribution, no problem. But if you have a large undecided block, and they go 80% for one candidate over the other, it could make a non-trivial difference in the final result.
Finally, for understanding the probability distribution within the bubble, the simplicity of my model doesn’t really allow for that. You would need to do something that took the polling information in the states and the uncertainties there, then run simulations that let you see the probability distribution on the electoral college level.
I don’t do anything like that (and probably won’t any time soon) but luckily there are lots of places that do:
http://horsesass.org/clinton-versus-trump-2015/
http://election.princeton.edu/todays-electoral-vote-histogram/
http://www.nytimes.com/interactive/2016/upshot/presidential-polls-forecast.html?_r=0
http://projects.fivethirtyeight.com/2016-election-forecast/?ex_cid=rrpromo
None of them really have a good display of the distribution as it changes over time, which a shaded bubble might give you, but they all give you a much better sense of the distribution of probabilities within the “bubble”. My very basic classification into states that are in play and states that are not essentially gives you a pretty wide range of possibility. Something would have to be very wrong with the polls in a systematic way, or there would have to be a very big last minute shift in November too late to catch in the polls in order to end up with a result outside of the bubble. But the method doesn’t give much insight into just how unlikely the edges of the bubble are compared to the center.
The tipping point doesn’t really directly map to anything with the bubble, other than if the tipping point is less than 5% it means that both candidates winning are within the bubble, and the sign of the tipping point tells you who is winning in the expected case. What it is showing you is how much of a uniform polling change across all states would be needed to change the outcome.
If you were going to do something else within the bubble itself, it would have to be something like showing where the bubble boundaries would be if the limit I chose for “close state” was varied to other numbers besides 5%. For instance, I could make a really dark green color for the range if I only counted close states at 1%, lighter for 2%, even lighter for 3%, etc. I think this would give you the kind of display you are suggesting Ben. And I do think it would look rather nice, and not be that hard to interpret. It sounds pretty cool actually.
Unfortunately, it would also be pretty hard to move from what I have now to that. I’d have to do a whole lot of things very differently to calculate and store all those additional values where the colors changed at the various percentage boundaries. So… not going to happen this year.
Maybe I’ll think about it for 2020. :-)
Ben,
Swing voting is not the same as undecided breaking.
A swing voter is one who says they’ll vote one way then changes their mind (probably most likely in the case of third party voters).
Undecided voters have stated no clear preference and could select any party. I believe there have been studies which suggest they break 2/3 to 1/3 in favour of the challenger (for non incumbent, I believe they go on party). In this election… Who knows?
Obviously in the first case because the vote changes it decreases one party in favour of another. In the second the change just benefits one party.
Conventional wisdom suggest that Libertarian and Green voters should swing to Republican and Democratic respectively and that undecided will break 2/3 to Trump, 1/3 to Clinton.
If that holds true, Trump can win without flipping any Clinton voters per se, he just needs to flip some Johnson voters, a majority of undecided and get out his supporters to the polls. All that without Clinton losing a single tenth of a point!
Is that 1 in 50? Easily, more like 1 in 10 potential voters changing their minds (be it from party to party or undecided to party)
I hope that makes some kind of sense, I don’t always explain quite as I see it in my head
Am I right in thinking this is what you’re describing with undecided and volatility?
fivethirtyeight.com/features/swing-voters-and-elastic-states/
Richard: Thanks for the explanation on “swing” versus “undecided”. I hadn’t known that before.
Abulsme: I keep thinking what I’m talking about shouldn’t require much extra calculation and certainly no simulations. There’s possibly something obvious I’m missing. I’ll try modeling it out by hand (slow, but if I’m making a mistake, I’ll be sure to see it.)
In the meantime, maybe I can simplify what I’m talking about even more. Instead of making a colourful, fuzzy graph over time, imagine a “valley” showing, just at this moment in time, how heavy of a lift each state is using the Y-axis. The graph would be a stair stepped ‘V’, with the point of the V resting on the X-axis, showing the expected electoral vote outcome. The states would be lined up on either side of the point in the reverse order as the State Breakdown by Category, but they would vary in width depending on number of electoral votes they carry. The Y-axis would be the absolute value of the Trump-Clinton margin for each state (in percentage points). To make the stair steps, each state has a vertical line going down, and a horizontal line going in the direction away from the expected result.
You could then imagine the state of the race as a boulder at the bottom of the valley. There’s a lot of inertia keeping it there, but enough energy could move it one way or the other. For example, Iowa and Georgia are both 0.8% above the floor, which we can (perhaps reasonably) view as equally heavy lifts. However, they have a different number of electoral votes, which means, once you lift that boulder 0.8%, if you’re pushing left, it’d roll 16 electoral votes (Georgia). Pushing right, it’d go only 6 ev (Iowa).
Does this make any sense? I feel like I’ve typed a thousand words when a single picture would have sufficed. Anyhow, if that is understandable, the fuzzy graph I’m talking about is simply replacing the Y-axis with shading and tracking the “valley” over time.
Sorry for being absent from many of these threads for, um, a month. Things have just gotten really busy with the volume of polls, so I’ve been pretty slammed. I’ll try to get to a few of these today.
On the suggestions for alternate visualizations, thank you! Always appreciated. I do kind of like what I have, but I’ll think about changes for 2020. It is too late for this year. :-)