Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

The Comey effect (original graph and post)

Hillary Clinton’s narrow loss to Donald Trump was influenced by many causes in the home stretch: complacency driven by conventional wisdom and polls (and yes, poll aggregation), which led to the media assumption that she would win, which in turn was a likely driver of the tone of coverage. And of course there is so much to say about the candidates themselves.

A month after Trump’s upset victory, the aftermath is still sending shocks through the United States and the world. In addition to a hard rightward move on policy, Trump, Pence, and Company appear to be bent on uprooting many institutions. The risk to the American system of government and life has been noted by both liberals and conservatives.

In mid-October, I said I didn’t think Trump would clear 240 electoral votes, a statement I paid for later by eating a bug on CNN. My error seems to be accounted for by two events: (1) undecided Republican voters coming home, and (2) FBI Director Jim Comey’s letter to Congress about Clinton’s email.

In the above graph of the Comey effect, each point shows the median margin for polls that were in the field on that day. Surveys that were in the field on the day of the Comey letter release, October 28th, were started as early as October 25th, which is where the “Comey effect” starts to appear in the graph.

As you can see, the immediate effect of Comey’s letter was a swing toward Trump of 4 percentage points, about half of which stuck. This was enough to swing Michigan, Pennsylvania, Florida, and Wisconsin. Many factors went into this year’s Presidental race, but on the home stretch, Comey’s letter was a critical factor in the final outcome.

Update. A commenter points out a complexity that I didn’t do a good job of explaining in the original post: the decrease in the Clinton-Trump margin starts at the graph’s “October 26th” – and the Comey letter was released on October 28th. The reason for this has to do with how I extract day-by-day information from multiday polls.

Polls that contributed to the “October 26” data point were also in the field on later dates, so that day’s median includes respondents who answered as late as November 1st. As we say in signal processing, it’s filtered in time. Conversely, people who responded on October 28th are included in data points as early as the one labeled “October 25th” – and indeed, that is when a change starts showing up. In short, the timing of the change is exactly consistent with the Comey letter. Readers are welcome to look at the data here (transcribed from Huffington Post).