01 November 2016

Statistikhengst's 2016 Electoral Landscape No. 2 (of 3): Clinton vs. Trump

Statistikhengst's Electoral Landscape No. 2:

Clinton vs. Trump
(Johnson, Stein, McMullin as sidenotes)

Data through October 31, 2016, 8 days before the election

Clinton 328 / Trump 210 / Margin: Clinton +118 EV
Clinton 288 / Trump 198 / Tossups 52 / Margin: Clinton +90 EV

This is my second electoral landscape prediction-output for the 2016 election. A lot of things have happened in my life, especially in 2016, and there was less time for detailed reports. Nonetheless, I have faithfully logged all of the polling data and also watched very carefully the events as they have passed. Making a report based on data ending at the end of the day 31 October 2016 is also beneficial as the Email debacle for Hillary Clinton was on last Friday and the initial polling, at least to date, is showing little change in the overall dynamic of the race - but that could change within this week, to be sure.

This report deals with the current statistics only. My feelings are 99.99% out of the way. It also does not factor in things like ground game, last minute advertisting, weather... none of that stuff. This is about the numbers, numbers, numbers...

First off is to note that this election, in terms of statistics, is radically different than either 2012, 2008 or 2004, which were all very clearly 2-way races. This time, in 2016, whether you want to call it a protest vote or note, former GOP Gov. of New Mexico and current Libertarian presidential nominee Gary Johnson had often gone into double digits in polling, enough for major pollsters to justify 2 sets of statistics: 2-way and either 3-or-4-way statistics, the fourth initially being Green Party nominee Jill Stein. However, Ms. Stein was also polling between 2-3% in many 2012 polls and came in on election night 2012 with 0.36% of the vote - and for this reason, I see not much of a Jill Stein factor in this election at all. However, a fourth party (Independent) candidate named Evan McMullin, from Utah, may play a crucial role in at least one state. He is on the ballot in 11 states.

So, what does this mean for the calculations? It means that I needed to rethink how to calculate polling averages, because only taking values from a 2-way race in a year that at least has a feel of somewhat being a 3-way race would be wrong. Visa-versa also applies. Therefore I have taken the mean average of the 2-way and more-way stats to assign aggregate numbers to the states. So far, so good. In most states that post both sets of stats, there is not a huge divergence, but in a handful, the difference between the two is very large - and will be explained later.

First and foremost, here is all of the raw polling data to date, in excel-table form:

2016 Pres Polling - BATTLEGROUND/LANDSCAPE Data 31 October, 2016

The tables include hyperlinks to every single poll, the survey size, the MoE and the topline data, with the margins automatically calculated. The tables are in reverse chronological order, with the newest poll at the top and the oldest at the bottom.

The very first tab in the document lists the number of polls per state. Now, this calculation goes all the way back to February of 2013 (yes, I have been collecting them since way back then...), so not all of the Clinton or GOP figures mean a specific Clinton-vs-Trump matchup. However, I am a real believer in publishing the ENTIRE data-stream and not just cherry-picking.

In 2012, a number of "safe" states were never polled. This time around, due to 50-state polling from Morning Consult (R), Ipsos/Reuters, UPI/Cvoter and even Google (which I do not use, the survey sizes for many states are ridiculously small and I refuse to log half-assed polls with MoEs of +10!), every state plus DC has been polled a number of times.

Based on the polling results, which, as mentioned above, you can find all in the EXCEL DOCUMENT mentioned above and linked HERE again,  Based on the current aggregate date, I come up with the following electoral landscapes, one without tossups and the other with tossups (+1 or less in aggregate margin):

More importantly is to note that, based on 125 polling matchups logged for 31 October, 2016, there are currently 12 states in the battleground zone:

That table looks like this on a map:

There are two types of pretty amazing symmetry to the above battleground data:

First, there are exactly 4 states on the GOP side of the tossups, but in the battlegrounds, there are 4 states on the DEM side of the tossups, but in the battlegrounds, and there are 4 states truly in the tossups, where no one should really be trying to make a solid prediction yet. 

Second, that makes for exactly 4 states in the Southwest, 4 in the Midwest and 4 in the South. So, the phenomenon is spread evenly throughout 3/4 of the nation, excluding the Northwest (big sky) and Breadbasket areas.

I bet you see a couple of surprises there, for instance, South Carolina. Well, two 50-state polls came in on 31 October and it has been more than a week since SC was polled by anyone else and indeed, IPSOS/Reuters shows the race at Trump +5 in a two-way, while GOOGLE 50 shows Trump +4 in a more-way, making for an aggregate of Trump +4.5.  I don't think anyone is really expecting Clinton to pull a surprise and win SC and likely, when more polling comes in, it may very well slip right out of the battlegrounds. 

But the same applies to Michigan, where the IPSOS/ Reuters shows Clinton +4 in a two-way but Google shows only Clinton +1 in a more-way, which when averaged with the others, makes for an aggregate of Clinton +4.68, which, like South Carolina, is on the outer edge of the battlegrounds. Plus, the two polls just days before in Michigan, Emerson (which has a mathematical conservative bias) and EPIC/MRA (one of the two gold standard polls for the state), show Clinton at +7.1 and +8 over Trump. So, discernment also plays a role, here.

This also means that states that have popped back into the upper edge of the battlegrounds may not stay there for long.

That being said, in some of the battlegrounds, we are seeing a huge amount of consistency.

In North Carolina, where early voting is well apace, with three new polls in the mix and only one out of the mix, it stands at Clinton +4 on the aggregate in the Tarheel State, and she has been around +4 quite often. This brings me to the point about Remington Research (R), the only pollster showing Trump ahead. Well, one week ago I noted that Remington's internals are weird at best and the Remington from last week was replaced by this one, so the effect on the overall statistic was pretty much nil. It is obvious to me that the Clinton team staked it's fortune on doing everything in their power to lock down Virginia and North Carolina in the hope of, regardless what may happen, effectively closing Donald Trump's path to 270.

In Ohio, on the other hand, more and more polling from different organizations is coming in showing Trump with the upper hand, and so, in spite of a lean Trump +2.05 in the state, the statistic is consistent as it is in North Carolina. If Clinton wins the national election, as the state polling statistics are indicating, but loses Ohio, then this will be the first time since 1960 that Ohio did not go with the winner and the state would lose it's bellwether status. Wait and see.

And so, were I to use some gut feeling, my instinct tells me that with consistent stats this late in the game, the first two state likely to switch sides are both NC and OH.

In Iowa and Nevada, it is more tied up, but each state has a "Clientele" more likely to fit the one nominee: Iowa (overwhelmingly white, very evangelical) for Trump and Nevada (large Latino and Mormon populations) for Clinton. Plus, early voting in Nevada, epecially from Clark and Washoe counties, is breaking records. But that means kind of nothing until all of the votes are counted.

On to Florida, where a recent spate of polls show Trump somewhat ahead, but the one St. Leo poll showed Clinton miles ahead. Now, sometimes polls come in waves for one side or the other. And in 2016, the aggregate showed Romney ahead by a nose the entire time, but yet, President Obama carried the state by +0.88%. Make no doubt about it: without Florida in the GOP column, Trump's path to 270 is absolutely foreclosed. Clinton, on the other hand, if she remains ahead in North Carolina and is practically 100% guaranteed to win Virginia, which is not even close to the battlegrounds anymore, adds 28 EV to the DEM column. Florida has 29 EV. So, the strength of VA/NC combined cancel out a possible Trump win in FL. Clinton is only ahead by +0.97 here, but compared to this time-frame four years ago, which showed Romney +0.76, Clinton is still running 1.73 points better now than Obama did then.

The one state that continues to surprise and which may really surprise on election day, is Arizona, an erstwhile GOP bastion.  However, the spate of polling in just the last four days has produced two Trump +2 values, a Clinton +2, Clinton +1 and a mathmatical tie, which puts Clinton just a nose ahead. Wait and see. This may just be smoke and wind when it comes to election night, but I don't expect to see the news networks calling AZ immediately, to be sure. Here, the Latino vote may be crucial.

Not many people are talking much about Pennsylvania, but a look at the values can explain why, also why more-way polling should actually be counted more than two-way polling:

So, in the last seven days, only one poll (the IPSOS) has produced a two way result, showing a far smaller lead for Clinton in the Keystone state than the other polls. But again, discernment is also important: the previous IPSOS showed her at +10, this at +2, an 8 point shift. And of course, Remington Research also shows a narrow margin. But F&M, one of the two gold-standard polls for this state, show +11. So, although the statistic shows Clinton +5.07 in the state, she is likely doing considerably better than this. And she is definitely ahead of Obama's statistic here 8 days away from the 2012 election. As is also the case in Wisconsin and and relatively similar to his statistic in Minnesota.

To see the rest, tab out to the EVERYTHING TABLE at the excel document above.

A key principle in watching election statistics is to remember that "a rising tide lifts all winning boats" which usually means that the winning side posts larger margins in its SAFE states, while the losing side posts leaner margins in its SAFE states. For a good part, those stats are playing out on the Clinton side, but in some states, they are mixed, more than people may want to admit. For this very reason, I made and update the PENUMBRA tab, which then moves the states from their 2012 order of partisan ranking to their partisan rankings based on the current 2016 polling aggregates. There are a lot of changes this year, in all directions. Here is the penumbra tab, first for Clinton, then for Trump. Both are in descending order of margins:

The two columns that show a difference, either to the actual 2012 results or to the end-polling aggregates for Obama and Romney, could be of interest to everyone. The more positive values, the more it means that Clinton and Trump have improved on the stats from back then. The more negative values, the opposite, of course.

On the DEM side, California, Massachusetts, New Jersey, North Carolina, Arizona and Virginia stand out as states where Clinton's current aggregate is far better than Obama's were going into the election. However, her current aggregate is far behind where Obama was in Vermont and somewhat behind in Delaware and Hawaii.

Trump is of course ahead of Romney's end-polling statistics in Iowa, Ohio and Nevada, but far behind where Romney was in Utah, Idaho, Alaska, and with less severity, in Tennessee, Texas and Arkansas.  

So, it's a mixed bag for both sides.

Before going to the national average, just a note again about Utah. No other state in the Union is seeing the situation that we see in Utah, where a fourth party candidate, Evan McMullin, really has a shot at winning the state. This could mean that the two-way statistics are worthless to begin with, because they compare Trump to Clinton and not to McMullin. And also because with such value mired in the 30 range, anything can happen.  I think we can look to the Beehive State to see that many Americans are very unhappy with the choice of both candidates in this election.

In the electoral landscape from one week ago, I wrote the following:

"I see four possible scenarios, none of them particularly good for Donald Trump:

Scenario 1: Trump goes and campaigns heavily in Utah (he was advised to do this and has not, to-date) in order to rally enough of the base to win in a 3.5 or 4-way race. This would mean less time for him in North Carolina, Ohio, Iowa and so on. Comparing 6 EV to 18? I bet he will go for the 18....

Scenario 2: Mitt Romney, Utah's favorite son, who has already openly said he will not vote for Trump or Clinton, comes out in support of McMullin, which would probably move the needle another 5 points or so for the former CIA-man.

Scenario 3: Gary Johnson, sensing that he has no chance, pulls out, and his supporters move toward McMullin,.

Scenario 4 (and I find this interesting): realizing that it is still unlikely for her to pick-up this ruby-RED state, but seeing an opportunity to deprive Donald Trump of Utah's 6 EV, the Clinton campaign could quietly, via surrogates, encourge the Clinton supporters to switch their voting preference to McMullin, which could move him to about 18 points ahead of Trump. This is a scenario that really COULD happen. There is nothing illegal or immoral about it, but it could be an excellent chess-move on the part of the Clinton campaign and I bet that their strategists are discussing this possibility right now, this weekend. Were I a strategist on her team, I would be doing exactly that."

Just to note: scenarios 1, 2 and 3 have not happened. And if scenario 4 were to happen, since it would be a whisper campaign....


The tracking list:

12 October 2016, Clinton +5.52, -0.21
13 October 2016, Clinton +6.61, +1.08*
14 October 2016, Clinton +6.75, +0.14
15 October 2016 - no battleground report posted
16 October 2016, Clinton +6,44, -0.31
17 October, 2016, Clinton +7.35, +0.92*
18 October, 2016, Clinton +6.91, -0.45
19 October 2016, Clinton +6.64, -0.27
20 October 2016 - no battleground report posted
21 October 2016, Clinton +6.74, +0.10
23 October 2016, Clinton +6.02, -0.72
24 October 2016, Clinton +6.25, +0.23
25 October 2016, Clinton +6.37, +0.23
26 October 2016, Clinton +6.73, +0.36
27 October 2016, Clinton +6.55, -0.18
29 October 2016, Clinton +5.91, -0.64
30 October 2016, Clinton +5.81, -0.10

31 October 2016, Clinton +5.93, +0,12
*Due to rounding, the difference is off by 0.01

So, three weeks later, we are now where we were on 12 October, 2016.
I will remind that a 6 point lead nationally practically guarantees an electoral college landslide. 

On this corresponding day in 2012the national aggregate was Romney +0.16%, so Clinton is currently +6.09 points ahead of where Obama was four years ago. On 10/30, she was +5,89 ahead, so we are seeing more stability here.

If you look at the margins, you see a gentle rising and falling - often called the sinus-curve effect. I fully expected to see this.

And some end-statistics: I first started collecting 2016 presidential polls at the beginning of October 2013 and since then I have logged 4,562 matchups from 2,443 individual polls. You can find the breakdown to this stuff at the EXCEL table, in the tab that says "poll totals".

For past reference, here is the ELECTORAL LANDSCAPE from 8 days before the GE 2012, from 10/29-30/2012.  And here is also the corresponding report from 10/26/2008.

One more thing: early voting has begun and there are lots of stats coming out of many states. Soon, I will be collating that data as well.


No comments:

Post a Comment

Constructive comments and critique are always welcome. Please keep it polite and respectful.