The 2016 Presidential election results came as a surprise to almost everyone involved - including both the Clinton and Trump campaigns, it appears. With shocking results like this, it is understandable that people would look for quick answers. Immediately after the election, analysts assumed this was a turnout issue, but it appears that overall turnout nationwide will end up being very close to or slightly above 2012.
While we have waited for the remaining ballots to be counted in places like CA, the reasoning has shifted to polling misses in key states that decided the election. Did the polls not anticipate new or very irregular voters turning out for Trump, or did previously Democratic voters finally make the switch en masse to the Republican ticket? Unfortunately, until the voter files are updated, the answer to that question is only speculation.
What we can assess right now is just when voters made up their minds about who to vote for and what the reasons were for their decisions. Using the cumulative data from Trendency Research from August until Election Day, we can track voters’ opinions in key swing states as it changed, enabling us to see just which moments were pivotal in deciding the outcome.
As a bit of background if you haven’t been following our op-eds this fall, Trendency Research is an online analytic platform where we convene panels of likely voters who log in regularly and answer questions on politics, current events, and other topics. Trendency looks at voter opinion on a continuum and not as a binary choice. Instead of choosing one candidate they support, Trendency surveys allow respondents to allocate their support on a sliding scale among candidates. This allows for an in-depth examination of support at various levels, which we term “Threshold Analysis.”
Voters for a candidate who are at higher Threshold are stronger in their convictions and less likely to shift to another person. Those at lower Thresholds divide their support among more candidates and more easily shift their allegiance. This can also be thought of in terms of volatility; those at higher Thresholds show less volatility while those at lower Thresholds show much more volatility in their level of support.
Two additional key measures we utilize, ones we will be examining in this piece, are the Commitment and Rejection Indices. The Commitment Index isolates those voters who are at the higher end of the Threshold spectrum and represent the core supporters for a candidate. These voters can be expected to vote for their choice on Election Day. The Rejection Index is the opposite of the Commitment as it measures those voters at the lower end of the Threshold spectrum who will likely never vote for that candidate.
We convened panels in ten states this cycle, but for this article we will focus our attention on five key states: Florida, North Carolina, Ohio, Pennsylvania, and Wisconsin. And while we reported data from these states individually in months leading up the election, we believe combining the data into a larger data set offers some interesting hints as to what were the most important factors this year.
After combining the data, we looked at both the Commitment and Rejection Indices for each state to see if we could discern a pattern. The charts below show the Commitment Index for each candidate in these five states:
From the chart above, we can see that Clinton’s Commitment Index levels generally followed a similar pattern, starting and ending in the same vicinity with a slight upward trend. A few states had more volatility in this measurement over time, but in general Clinton started in the mid to upper 30s (an average of 35.6) and ended up slightly above or slightly below 40 (and average of 39.8). In all five states her net movement was positive, although only in two states did she finish the election at a high watermark – Ohio and Florida.
For Trump on the other hand, his Commitment Indices fell into two camps. In North Carolina and Florida, he started and ended with a Commitment Index over 40, which is quite solid. In the Rust Belt states (Ohio, Pennsylvania, and Wisconsin), Trump had some work to do, with an Index that started in the low-30s. He clearly improved on those numbers, however, and in the end his Commitment Indices rose in the Rust Belt, as would be expected based on election results. Pennsylvania was the only state that remained on its own, in terms of the Commitment Index.
Interestingly, Trump ended up where he started in Florida and North Carolina, while the movement in the Rust Belt states was clear (especially in Wisconsin). Similar to Clinton, Trump ended on Election Day at a high watermark in just two of the states.
If we take these states and look at the average across them, a very clear trend shows up for both candidates.
As the chart above shows, Clinton and Trump were basically neck and neck in support until Trump opened a clear lead right before the first debate. And while the debates didn’t drop the level of Trump’s core support too much (about 1 point a couple days after the first debate), the debates clearly helped Clinton. From right before the first debate through late October, her average Commitment Index increased from about 37 to 40. Trump’s average Commitment Index, on the other hand, started its upward trend right before the third debate and moved from 38 to 41 right before Election Day.
Another big takeaway from looking at the Commitment Index is that the release of the Comey Letter announcing the FBI was looking into additional emails clearly stalled the momentum Clinton had going. That being said, Trump had already caught up to her right when the letter was released. Interestingly, his upward momentum also stalled for a couple days, but then shot up again starting on November 2nd, a few days before Comey announced there was nothing new in the emails.
This is the view from the positive side of the equation, but looking at the Commitment Index alone doesn’t tell the whole story. To do that, we need to flip around and examine the changes in the Rejection Indices.
Clinton’s Rejection Indices remained relatively stable throughout the fall, except for in North Carolina, where she started with a much higher Index but showed impressive movement until late October. During the last two weeks before the election, we see an upward trend in most states.
From August until Election Day, Clinton was able to lower her Rejection score in three of the states, while in Wisconsin and Ohio her score had a net increase. In even worse news for Clinton, she ended up with a score of 50 or higher in all five states. Normally this would have meant a blow out loss for her in all five states, but as we know 2016 was anything but normal.
Trump’s Indices were more scattered but generally followed a gentle upward slope throughout the campaign, until the last two weeks when there were differing views among voters in the five states. Unlike Clinton, whose Rejection scores went up over the last two weeks in four states, Trump’s score only went up in two states in the last two weeks. He remained flat in two, and his score dropped sharply in Ohio.
Interestingly, we did not see the same herding pattern for Trump on the Rejection Index that we did in the Commitment Index. Voters in Florida began in different places, and while they ended up very close there were very distinct paths to get there. Another key piece of data was the fact that Trump was able to improve his Rejection Index in one state over time, and that was Wisconsin. In the others, he ended up in a worse place than he started.
If we look at the averages, we once again see very interesting patterns.
Among these five critical swing states, Clinton held a higher average Rejection Index from mid-August all the way through the first debate. Immediately after the first debate, however, her average Index dropped two points while Trump’s starting drifting upwards. From that moment until the day the Comey letter was released on October 28th, Clinton held an advantage over Trump. It appears that the unprecedented release of that letter by the FBI Director had a small but important negative impact on Clinton. It also helped improve Trump’s score, but only for a couple days when the two candidates’ Rejection Indices both increased at the same rate on average.
This same impact is even more apparent when we look at the net average (Commitment Index minus Rejection Index):
On the above chart, the higher the line the better for a candidate, as this means that their Commitment Index is higher in relation to their Rejection Index. Ideally, a candidate’s pool of committed voters is larger than those who outright rejection them, but that hasn’t usually been the case for most elections we have covered and especially not this year with Clinton and Trump.
Here we see that the first debate totally altered the landscape of the election. Trump had been leading in these states on this net index scale through August and most of September, but once he and Clinton faced off on the same stage at the first debate and the subsequent Access Hollywood video was released, her numbers shot up while Trump’s dropped.
Almost as soon as the debates ended and Trump wasn’t in front of a live TV audience speaking, his numbers began to creep up again. At the same time, right when the Comey letter went public, Clinton’s net difference index took a plunge, while Trump’s continued up. He eventually passed her six days before the election, where his numbers flat lined, but Clinton’s continued to sink.
So, did the Comey letter cost Clinton the election? Not in and of itself, but it clearly had a direct and immediate negative impact on Clinton. Trump’s numbers were already on the rise, though, so there is no guarantee that he would not have won regardless. Also, Clinton’s numbers had stalled for about a week before the letter and the two lines were on a path to meet by Election Day, if the pattern had held. Clearly we will never know what it would have looked like, but there is not a simple or declarative answer to the Comey question.
The American public might have gouged their eyes and ears out if there had been a fourth debate, but more debates probably would have helped Clinton, since those were nothing but a success for her and increased Trump’s negative numbers. As we mentioned in the beginning of this piece, once we have all the data from this election appended to the voter file, analysts will be able to determine the true makeup of the electorate. However, that still won’t answer the crucial questions about what moment may have led to white voters in rural Rust Belt states to jump to the Trump camp.
This Trendency data does shine a light, however, on those key moments - the first debate, the ending of the final debate, and the Comey letter - and how those impacted and ultimately determined the outcome of the election.