Scandals' Impact on Polls: A User's Guide
On Friday, FBI Director James Comey sent a letter to Congress saying he had found new emails that appear to be pertinent to the previously closed investigation of Hillary Clinton’s private email server. This type of news -- potentially poll-shifting events that happen with barely enough time for surveys to gauge their effect heading into Election Day -- is not particularly easy for quantitative elections analysts like me to deal with. It will likely take at least a couple more days to get a complete picture of what impact (if any) this new controversy will have. And if the Clinton campaign releases anti-Donald Trump opposition research this week, it’ll be even more difficult to figure out how public opinion will shift before next Tuesday.
When events like this happen, it’s often helpful to take a look at the basic components of the race before getting too bogged down in the details. I attempted to do that using a very simple model (described below) based on the idea that the candidate who is disliked the least on a given day will probably have a polling advantage. The model suggests that if this new email scandal drags Clinton down in the way past controversies have, the race could end up being very close. But if it doesn’t (or, again, if Clinton were to reveal damaging information about Trump), she may maintain her polling advantage heading into Election Day.
Simple Model: May the Least Disliked Candidate Win
The bird’s eye view of this race is relatively simple -- there’s no major war, massive recession, economic boom or overwhelmingly popular (or unpopular) sitting president shaping the election, so whichever candidate is the subject of negative press coverage tends to lose ground and give the other candidate the chance to grow his or her support.
For example, when Trump got into a public fight with the Muslim parents of a slain American soldier, his favorability rating went down and he dropped in the horse race polls. Similarly, when Clinton faced scrutiny for her email practices and health controversies in September, her poll numbers took a hit.
I translated that intuition into a simple model that predicts Clinton’s or Trump’s lead in the four-way RealClearPolitics national poll average (which includes Libertarian candidate Gary Johnson and Green Party candidate Jill Stein) using their favorability ratings on that day. There are a number of limitations to this approach (e.g., it extrapolates to values not covered by the input data and would probably be better modeled as a time series). But the model explains about three quarters of what’s going on in the national horse race and intuitively makes sense, so it’s worth thinking about. The results are shown below:
The table tells a relatively simple story. If both candidates have the same low net favorability rating (the percentage of voters who view a candidate unfavorably subtracted from the percentage who view him or her favorably) the race should be close. If one candidate seriously outpaces the other in favorability -- as Clinton has been doing for significant parts of the fall -- that candidate builds a significant lead. And if both candidates were as well-liked as traditional candidates (i.e., with a zero net favorability rating) the model predicts a three-point Republican win. That matches what Alan Abramowitz’s Time for Change model (which forecasts solely on economic and political considerations and assumes two roughly generic candidates) predicts. (Note that colors in the table reflect the level of lead each candidate has: Redder cells indicate a larger Trump lead; bluer cells show a larger lead for Clinton; and lighter cells show a close race.)
What It Means
The model suggests that if the current email news sticks to Clinton in the way past controversies have, the election might be close. In July and September (months where Clinton faced scrutiny from the press over her emails and health) her net favorability rating was often between -13 and -17 points. Trump’s current net favorability is -21.0 points, according to the RCP average. If Trump sits at a -20 net favorable rating and Clinton ends up around -15 points, then the model predicts her lead would diminish to about 1.4 points.
That’s not a particularly safe lead. In that scenario, some small errors in this (very simple) model or a small- to medium-sized polling miss in Trump’s favor could give the GOP nominee the White House. And if both candidates end up with a low, roughly equal net favorability rating, the race could be virtually tied.
That being said, we don’t yet know if this story will stick to Clinton through Election Day. If Anthony Weiner (whose laptop was being scrutinized as part of the separate investigation that nonetheless triggered the new one into Clinton’s email) takes most of the attention or if Americans shrug in response to it, her net favorability might stay closer to -10 points. In that case, Trump would need to suddenly earn higher favorability ratings than he’s had for the entirety of this race to equal Clinton’s standing in the polls.
It’s also unclear if the Clinton campaign has more opposition research ready to drop on Trump. If so, it’s conceivable that Clinton could have a -15 net approval rating and Trump could slip back to -25 points. That would translate to a 3.4 point Clinton lead on Election Day, despite Americans not having a great affinity for either candidate.
These scenarios are admittedly speculative, but the point of this model isn’t to predict the future. The point is to provide a data-driven anchor for readers. Polls may not clearly show the effect of this controversy for another couple of days. And if some other controversy emerges on Wednesday or Thursday, there may not be enough time for polls to fully gauge its effect by Election Day. This model could then help readers keep their bearings and make educated guesses as how late-breaking news might change the state of the race.