The key factor in any election is almost always who will turn out to vote.

Pollsters usually have two categories of potential voters with which to measure anticipated voter sentiment.  The first is registered voters (RVs), and the second is likely voters (LVs).  The former are easy to identify, but since so many RVs don't actually vote, a poll of these persons is relatively useless once an actual campaign begins in earnest.  LVs are much more difficult to identify and require a number of subjective assumptions by pollsters.  One of those assumptions is based on whether or not someone voted in the previous election.  A second is what the person being polled tells the pollster.  It takes many more inquiries to find a likely voter, and thus many pollsters prefer to employ registered voters as their sample.

At this point in the 2016 national election cycle, and particularly in the presidential contest, a poll based on RVs is virtually useless.

Historically, a poll based on carefully chosen LVs is much more likely to be accurate, although even these polls have limited utility until just before election day itself.  This is partly because there are usually a large number of undecided voters, and this number diminishes as the election approaches.

In 2016, the value of LVs is even less than usual because of an important new circumstance.  That circumstance is a consequence of what I have called the "mutiny of the voting masses" in the nation.  Although quantifying the voters who usually do not vote but will likely vote this year is difficult, the reality that there will likely be a very high turnout of these voters in 2016 is not speculation.  An examination of the results of the caucuses and primaries in the past months leading to the two major party conventions is irrefutable evidence that there is a new category: what I call "new likely voters" (NLVs).

NLVs are obviously more difficult to identify than ordinary LVs.  For pollsters, it means more time and more expense.  Private (and almost always unpublished) polls for individual candidates and political parties are paid for with the understanding that they will be relatively accurate.  Pollsters who do this kind of paid polling might be expected to spend this time and expense, especially because if they are inaccurate, they will soon be out of business.  But the polls most of us see, those done by media outlets and major public polling firms, will likely not spend the time and money in 2016 to identify NLVs.  Instead, they will rely on LVs – and hope for the best.

That hope is not likely to be fulfilled, I suggest, because there will probably be so many voters on November 8 who have rarely if ever voted before but are motivated to go to the polls this year.

A lot of anticipation, even assuming this is true, remains as speculation.  Traditional demographic electoral analysis often breaks down voters into ethnic, religious, gender, and racial categories.  It is not only the percentage of these groups that a candidate might receive, but the numerical turnout.  For example, the Democratic nominee could be expected to receive approximately 90% of blacks who actually vote, and that might not change much from recent previous elections, but what percentage of the total of black voters will actually turn out?  Is it reasonable to think Hillary Clinton will draw a black turnout similar to the turnout that Barack Obama drew?  In the case of Hispanic and Jewish voters, there is some evidence that the percentages who will vote for the Democrat or the Republican might actually change somewhat.  In the recent past, more women have voted than men, but what if in 2016 there are more male voters than women who go to the polls?  I suggest that the likely turnout of NLVs significantly upsets the polling models we have accepted in the past.

I have lived through an election such as this one might be, albeit on the state level.  Even those who do not live and vote in Minnesota might well remember the extraordinary upset election of non-traditional politician Jesse Ventura in 1998.  A month before the election, every poll had him in single digits against two well known major party candidates.  Even very late polls indicated only that he was gaining substantially, but not that he would win.  In fact, this election had a major influx of NLVs – virtually all models of voter turnout were broken.

For several weeks, both national and state polls have indicated that Mrs. Clinton has a growing lead over Mr. Trump.  Most recent polls have indicated that this lead has narrowed notably.  In at least one major poll, Mr. Trump is now ahead, and in others, the race is a virtual tie.  Even assuming that most polls are now composed of LVs, how many of them are accurately polling NLVs?

Many of my colleagues are depending on the polls to analyze this election, particularly the presidential election.  A number of them have already concluded that the election is over and that the only question is about the size of Hillary Clinton's victory.  She might indeed win, and might even win by a big margin, but I suggest that this is not proven by any polls we now have.

In fact, if my contention about "new likely voters" is accurate, and barring the unforeseen (always possible in a cycle like this one), I think the voting patterns of 2016 indicate that the election could be heading for a historic upset of the political assumptions and models most of us now use.