Late polling in Florida, North Carolina and Nevada probably misses some Trump support

Because of the problem that early voting presents for pollsters, the polls showing a dead heat in early voting states may actually reflect a Trump lead. This problem stems from the problem of identifying who is likely to vote.

Every poll represents a best estimate based on limited information

Because voters will label themselves likely to vote most of the time, pollsters use a variety of methods to ferret out who will actually vote. Examples of methods used to find likely voters include assessing the voting history of a poll respondent by gaging the ability to answer basic questions about the past election or merely asking them if they voted. Other methods exist, but for the most part it's a straightforward assessment.

Inevitably some respondents who receive the designation as a likely voter will not actually turn out and vote. Unforeseen events intervene or perhaps the person simply forgets to vote. Statistically such events are inevitable. This sort of thing represents one reason why pollsters employ margins of error. Of the flip side, some voters who fail the likely voter screen will actually vote. This is the source of the bump Democrats receive.

Increasingly, one can walk in and vote early with ease

The bump is illusory, just a function of which side voted first

Republicans vote on election day, so their "unlikely" voters are less likely to get banked in pre-election polls

The military vote is one portion of the early vote that goes to the GOP, but overseas soldiers are not polled by phone

Our election cycle lasts for about two months now. First, because this is true, you have a wide range of time within which people vote. Second, because in every poll some people who are deemed unlikely to vote actually end up voting, you will have people included in late polling that were not included before.

In other words, a person who pollsters deemed unlikely to vote one month before the election because he or she failed a likely voter screen will be deemed a likely voter after they have voted. They no longer fail the likely voter screen because now, unlike before, they have recently voted. This particular problem in polling advantages Democrats because they tend to vote early.

Democrats always do everything they can to get their voters out to vote before election day. Republicans always win on the actual election day itself. The question becomes whether Republicans will vote in enough numbers to offset the early Democratic vote.

This problem will produce a late surge for Democrats in polling because they are banking more of their unlikely vote than Republicans. Because Republicans deemed unlikely to vote, but who will actually vote, do not do so until election day, that vote goes uncounted in polling when the Democratic unlikely voter who actually votes will get counted.

So the long election period that we now see in most states makes polling hard. Two states where early voting is not allowed nearly as much as other states include Pennsylvania and New Hampshire. Because of this, we can probably assume the polls in those two states will more accurately reflect Trump's level of support.