Debate Night: Undecided is Everything, Advantage Trump

Likely the best data driven write-up of the state of the race between Hillary Clinton and Donald Trump heading into the first debate tonight is Nate Cohn’s over at the Upshot. Cohn gives two theories of the race. On one, Clinton is really still quite a bit ahead and polling saying other wise is mostly just a snapshot of noise that will go away as election day nears. The other, and the one that makes more sense of the date I’ve collected, is that the race has in fact substantially tightened, likely into a virtual tie. Clinton’s missteps and Trump’s relatively good behaviour, especially since mid-August when Kellyanne Conway took over as his campaign manager, have erased what was a 6.5% lead on August 9th.

The data that I’ve been keeping assiduously since late August says it’s the latter theory. United Press International was the first to note that how well Trump controlled his tongue has a massive impact on the horse race. My numbers suggest that is the best way to look at things. When Trump re-brought up birtherism, suggested Clinton’s Secret Service detail stop carrying guns and see what happens, and Donald Trump Jr. tweeted out a ridiculous Skittles meme within the space of a few days, Trump’s numbers took a noticeable dive beginning early last week. By the end of the week, and into early this week, Trump made no major blunders and his numbers have rebounded back into a tie or a half point lead for Clinton in polls in the field in the last few days.

One reasonably reliable set of polling numbers suggests that as many as a 1/3 of voters will substantially determine how they vote based on the debates, beginning tonight. Meanwhile, somewhere between 15-25% of likely voters at present are undecided, planning to vote for Gary Johnson or Jill Stein, or have a soft commitment to their current candidate of choice. All of this makes it less than surprising that Clinton’s national lead is hovering between 1-2%, and there are as many as 15 states currently up for grabs. My #10at10 accounting posted daily on Twitter – all national polls in the last 10 days at 10 a.m. – currently sits at 1.6% ahead of tomorrow morning, but Trump briefly took the lead last weekend before his tongue troubles began impacting the polls. Meanwhile, by the same measurement (all polls completing their field work within the last ten days, no adjustments), Arizona, Colorado, Florida, Georgia, Iowa, Maine, Michigan, Nevada, New Mexico, North Carolina, Ohio, Pennsylvania, Virginia, and Wisconsin all see Clinton and Trump within 5.0% points of each other in terms of spread. FiveThirtyEight sees New Hampshire as currently a 2.2% race, while #10at10 figures would peg it much higher for Clinton at a 7.4% gap. Another 4-5 states are somewhere between a 5% and 10% race currently.

Here’s what the numbers look like in graph form:

Nationally

10at10-sept-27-night-before

States

states-10at10-debate-night

States Numerical Chart

screen-shot-2016-09-26-at-7-35-52-pm

The tipping point state, the state that would determine who wins if the electoral race is otherwise nearly tied, looks to be Michigan, Colorado, Pennsylvania, or Florida. As of now, Clinton is ahead by 1.0% to 2.2% in all four (neither candidate can likely win without winning at least two of those states). While most poll aggregators still have Trump up in Florida, the very most recent data suggests Clinton has taken a slight lead there while dropping steadily in Colorado, Michigan, Pennsylvania and elsewhere.

The major difference between this way of doing things and all other poll aggregators is that it insists on using virtually all polls (Google 50-state numbers with under 100 for a sample size are totally excluded) while not adjusting them. FiveThirtyEight includes virtually all polls, but adjusts them before its national and state projections. Huffington Post (and following them NYT/Upshot and Sam Wang at the Princeton Election Consortium) and Real Clear Politics exclude many polls. Huffington Post’s criteria much more clear and mostly consistent.

By using just 10 days of data, this model is much more sensitive to quick changes than even FiveThirtyEight’s “Now-Cast.” How soon will we start knowing who “won” the debate in terms of its effect on reputable polling? While the L.A. Times polling is almost certainly consistently biased in favor of Trump by 4-6 points, its methodology has proven very good at consistently capturing the direction of the race before any others. Their numbers out about 4 a.m. on Wednesday, followed by quickly by other daily tracking polls, will give us a sense whether the first debate is going to move the needle much, if any.

Clinton is a known quantity and has a fairly strict ceiling, most likely, of about 44-46% nationally. Trump’s tongue wagging is key to the race. A few errant comments in the debate or the weeks ahead and Clinton could wind up with 350-400 electoral votes. But Clinton does not seem to control her own destiny. An October surprise from Wikileaks or Trump managing his verbal outbursts for another six or seven weeks could very well see him inaugurated next January 20th. Of course, all this assumes that both sides leave the voting machines to do their thing without interference (not at all a totally justified assumption).

Doug Johnson Hatlem writes on polling, elections data, and politics. For questions, comments, or to inquire about syndicating this weekly column for the 2020 cycle in your outlet, he can be contacted on Twitter @djjohnso (DMs open) or at djjohnso@yahoo.com (subject line #10at10 Election Column).