Polling Benchmarks Trump Needs to Overtake Clinton
Can Donald Trump win this election? And if a comeback is in store, when will the polls start to show it?
It's impossible to know if Trump will make a comeback. Clinton is the clear favorite and leads in the polls, but there are still more than two months left in the race. It's possible to imagine scenarios where Trump catches up and overtakes Clinton.
But it's important to try to figure out if those scenarios are plausible and what the early signs of a Trump comeback might be. To that end, I analyzed polling from 11 presidential elections and came up with three rules of thumb - Trump can still win, to do so he needs to be within a few points in late September, and there isn't much time to move polls in October - to help guide readers through the next two-plus months of polling. These guidelines came from an analysis of historical polling data, which is summarized in this graph:
The graph shows the days until the election on the horizontal axis. The vertical axis shows how far those polls were off from the final polling average heading into the election. The black line shows the trend.
The data (skip this paragraph if you’re not interested in the details) was provided by Professors Christopher Wlezien and Robert Erikson, the authors of the highly regarded book “The Timeline of Presidential Elections.” Erikson and Wlezien collected polls and (using interpolation and averaging polls on the same day, which differs from RCP’s poll averaging methods) calculated estimates of national presidential polling from 1952 to 2012. This graphic shows the difference between their estimates of the polls on a given day for the final 60 days of presidential elections and a final poll average (the three days leading up to the election). Polls taken before convention bounces had faded (three weeks after that year’s final party convention or earlier) were omitted, as were polls in elections before 1972 (I made a judgment call that polling was too sparse for our purposes prior to that year). The line shown is a local regression -- the span was chosen via cross-validation.
A few quick caveats before getting started. This graphic isn’t meant to explain everything about the movement of polling in the final months of the campaign. It doesn’t predict which candidate will win or what the winner’s margin might be. And the trend line explains a relatively small amount of what’s going in the data. The point here isn’t to make specific predictions or draw firm conclusions but to give a few general guidelines that can help readers interpret presidential polling for the next few months.
Trump Could Still Win
Right now, there are still 70 days until Election Day, and Clinton is up by six points in the RCP average. If Trump holds steady, he’ll be below the roughly seven or eight points that an election might move between September (about 60 days from Election Day) and November. In other words, he still could make up the gap between himself and Clinton. As FiveThirtyEight’s Harry Enten pointed out, Gerald Ford, Ronald Reagan and Hubert Humphrey (in 1976, 1980 and 1968 respectively) improved their post-convention standings by a larger margin than what Trump would need to win.
It’s also important to note that this graph might overstate the amount polls could move in early September. In recent elections, the electorate has become more polarized and polls have moved less, but some of that data was omitted because of the effect of late conventions. Despite that issue, the gap between Trump and Clinton is still small enough that it could close before Election Day.
Trump Should Try to Get Within Three or Four Points by Late September
The graph also shows that in past elections, polls typically don’t move more than four points after late September or early October (about 30 or 40 days before the Election). It’s pretty easy to see this in the graphic -- just compare the density of points under the trend line to those above it. So if Trump wants to be in striking distance of Clinton, his best bet is to close the gap to three or four points before October.
It’s important to note this isn’t an ironclad rule. If this election follows the recent trend -- a polarized electorate with relatively few swing voters -- then it will be difficult for Trump to move the polls more than a couple points. But if this election follows an earlier historical pattern -- with less polarization and more voters remaining undecided -- he may be able to make up more ground later.
There’s Not Much Time to Move Polls in October
Once October hits, Trump would need to move opinion quickly in order to win the race. Roughly three weeks out from Election Day, the polls start to converge reasonably quickly on their final levels. This makes sense intuitively -- Trump and Clinton will have already debated once and many voters will have learned about both candidates and made their final decision. So that final downward dip shows that undecided voters don’t always wait until Election Day -- many of them settle on a candidate before or during the month of October.
One could argue that the downward dip in the graphic is an artifact of the math involved -- that is, the final polling average contains the polls from the final three days, so the final points will inevitably be close to zero. But when other metrics were used (e.g. difference between polls and final results) the trend line had the same general shape.
What This Graph Misses
There are also some sources of error and uncertainty here outside of the data.
First, these are national rather than state polls. In most models, state polls provide more information than national polls. Heading into the 2012 election, national polls showed a closer race than state polls, but the state polls ultimately proved to be more accurate. So these polls are far from the only data source on public opinion.
Second, a polling miss is possible. Essentially, if pollsters incorrectly predict what the electorate will look like or fail to catch late movement in the polls, they can accidentally miss the election results by a few points. In 2014, pollsters underestimated Republican strength and in 2006 they underestimated Democratic strength. So if Trump or Clinton pulls a little ahead in the polls, this sort of error could still allow the other candidate to pull out a win.
Finally, the fact that Trump can make a comeback in no way mean that he will. These findings also suggest that Clinton could turn her lead into a landslide. It’s possible to imagine a scenario where Trump continues to make missteps, Clinton finds better footing on questions about trustworthiness and her emails, and the GOP spend most of its money on saving the Senate. The result: Trump loses by a large margin.
But we’re not there yet. Trump still has time to right the ship -- or maybe sink it in historic fashion.