The election might be crazy, but the polling numbers aren’t

It looks like this absurd and lurid presidential election will remain unpredictable until the end. Between the FBI's on-again, off-again investigations of Democratic candidate Hillary Clinton's private email server, the "you can do anything" comments about women from Republican rival Donald Trump — not to mention the unexpected injection of Anthony Weiner's sexting habits — it's hardly surprising that the polls seem to show wild swings in voters' views.

Our experience as pollsters, though, shows this isn't so. It's true that daily tracking surveys reflect dizzying twists and turns in the support for Trump and Clinton. If you compare the single daily tracking poll from Reuters/Ipsos with the Huffington Post/Pollster poll aggregator, you'll see both show what looks like lots of movement in support for both candidates. Over five-day averages measured from July to October, the number of Trump backers rises from a little over 30 percent to around 40 percent; Clinton's support fluctuates between 40 percent to almost 50.

However, these momentary swings disappear when we instead look at monthly averages. The larger slices of time show that the rapid swings in voters' views always return to a rough equilibrium. In fact, there has not been any real change in Trump's and Clinton's relative position over the past three months. In July, Clinton had a 4 percentage point lead over Trump; in October, she still had a 4 percentage point lead. The most recent polls show that her lead may have narrowed since FBI director James Comey announced last Friday that the FBI is investigating more emails as part of a probe into Clinton's use of a private server, but our experience suggests that this gap may widen again.

Why has support for the candidates been so stable? The reason is that most Americans already know – and have known for a while — who they support. Despite this election's characterization as one shaped by "swing-ier voters," the race has pretty much been on cruise control for most Americans. At the start of the year, an overwhelming 99 percent of Americans had heard of Clinton, and 98 percent of Trump. Only 24 percent of Americans expressed neutral opinions on the Democratic candidate, and 23 percent on the Republican. Voters have been exposed to Clinton and Trump, in one form or another, for more than 30 years; they knew what they thought of them.

What the daily polls actually show is the waxing and waning of enthusiasm among supporters of the rival presidential candidates. They're doing this in two ways. First, through use of "likely voter" filters. Most likely-voter filters, including our Reuters/Ipsos one, include a measure of voter enthusiasm (along with things like registration and past behavior) as a factor in determining whether the respondent will turn out to vote on November 8. As their enthusiasm increases or decreases, respondents are more or less likely to be classified as a likely voter and therefore reported in our "horserace" polling. When events excite or depress the base, they show in our polling. Our five-day rolling average of enthusiasm to vote among Clinton and Trump supporters since July shows that shifts in likely turnout tend to correlate with shifts in the polls.

The second way polls show shifts in enthusiasm is (potentially) through "non-response bias." Non-response bias is the idea that groups of people may be so discouraged by the events of the campaign that they do not answer when contacted by pollsters. We say potentially because this is hard to measure directly — we cannot survey the people unwilling to be surveyed. But research has shown that this could have a significant impact on temporary poll shifts.

However, we can tease out some of this impact by observing the stated party preferences of our poll respondents over time.The Reuters/Ipsos poll does not normally sample, quota or weight the partisan identification of its respondents. Because party identification is subjective — a personal stated allegiance rather than an objective demographic — there are no commonly accepted benchmarks for the "right" party composition of the electorate. That means we ask people how they identify, and they answer as they answer. However, the partisan composition of our poll — and most other polls — can fluctuate substantially over time. Since July, the Reuters/Ipsos polls show that as scandals pop up, people with strong party affiliation are less likely to respond to a poll.

This is far from a "flat" line, and it underlines the extent to which these identities are variable, based on political dynamics and events.

We need to keep all this in perspective in the final week of the campaign. Observers need to distinguish between whether voters are switching candidates or simply becoming more excited about their candidate of choice. For the candidates themselves, it means the sprint to the finish is about keeping their supporters fired up while demoralizing the other side's base.

Chris Jackson is a vice president at Ipsos and runs the Reuters/Ipsos poll. Alanna Spurlock is a research analyst at Ipsos.