Why Trump just got his best two polls of the cycle

Take the polls in a state like Florida. It may seem almost impossible that last week we had Monmouth University showing Biden up by 5 points among likely voters (high turnout model) and this week we see a 4-point spread in Trump's favor among likely voters in the ABC News/Washington Post poll.

Of course, each of these polls come with a sampling margin of error. Both polls had reported margins of error of a little less than +/- 5 points. That margin of error, though, applies to each of the candidates, not the margin between them. For the ABC News/Washington Post poll, this means that 95% of the time (i.e. the 95% confidence interval), Trump's true percentage is somewhere between about 46% and 56%.

A quick rule of thumb is you need to double the reported margin of error to understand what the 95% confidence interval is for the spread between the candidates.

In the case of the Monmouth poll, Biden's 5-point lead could realistically be a deficit of around 4 points or a lead in the double-digits. Likewise, the ABC News/Washington Post poll could realistically be a lead for Trump in the double digits or a Biden advantage of around 4 points.

It's actually a good thing to see a wide spread in the polls. That means pollsters are doing their jobs. If we weren't seeing such differences, then it might mean that pollsters are putting their thumbs on the scale because they were afraid to release a seemingly outlier result.

Usually, the actual state of the race is in-between what the best polls for one candidate are and the best polls for the other candidate.

In Florida, the average has Biden leading by around 2 points.

Likewise, in Arizona, Biden's probably ahead. Last week, Biden was up 9 points among likely voters in a New York Times/Siena College poll. That's quite different from the ABC News/Washington Post poll in the Grand Canyon state.

Such a spread is, as in the case of Florida, a signal that pollsters are doing their job.

The average of all the Arizona polls has Biden up by around 4 to 5 points.

We look to the average because it's most often closest to the truth. Chances are Biden is leading in both Arizona and Florida.

Still, sometimes the average is wrong. If you go back and look at close presidential races (i.e. those where the final polling average was within 10 points) since 1972, the polling aggregate on Election Day misses by an average of a little more than 3 points.

Biden would likely still be ahead at this point in the Electoral College if every state was biased toward him by 3 points. Although Florida would fall to Trump, he could accumulate a majority of electoral votes in other states.

Since 1972, the 95% confidence interval for polling averages in close races has been closer to +/- 9 points. I call this confidence interval the "true margin of error" because it takes into account more than just sampling error. Other errors include the potential for the types of voters turning out being different than what pollsters anticipate (e.g. more young people than expected casting a ballot). Another potential cause for a pollster misfire is that one candidate's supporters are less willing to answer polls.

The bottom line is that there are a slew of errors that the traditional reported margin of error doesn't take into account.

The true margin of error could obviously make it so either candidate does better than anticipated. The polls may be underestimating Biden and he may be headed to a blowout.

The polls though could also be underestimating Trump. Biden's advantage in the states that will likely determine whether he wins or loses the election is only about 5 points.

History tells us that Trump could close a 5-point gap by Election Day. But let's say the polls don't shift from now until the election.

Even if Biden maintains his current lead, past errors indicate Trump will still have a non-nominal chance to pull off the victory.