"The dam is about to break," Pat Caddell, pollster to Jimmy Carter turned frequent Fox News commentator, declared earlier this week, projecting that recent polling trends in the presidential election which have been heading in Donald Trump's direction were about to snowball into a Trump landslide.
Caddell is one of the most prominent critics of the public polling averages and the media polls that are fed into these averages which suggest that Hillary Clinton holds a lead in the race.
But Caddell is not alone; a chorus of voices all the way up to the top of the Republican ticket have amplified the cries that the polls are skewed, or rigged. The perception that the polls missed the 2012 election badly, missed the Brexit vote result in the U.K. and missed Trump's rise have all fed the notion that the polls have long been failing to reflect the true will of the people, and that elites and numbers nerds are in for a great big surprise come Nov. 8.
The ABC/Washington Post tracking poll, which has an exceptional track record at predicting the national presidential vote margin, has shown a precipitous collapse of support for Clinton in the last two weeks, taking her from a 12-point lead to a tied race as of Wednesday.
Voices that one week ago decried the polls as "skewed!" now champion them as proof of a Trump surge; those who dismissed critics one week ago now find themselves sifting through poll samples seeking evidence that Trump is not really tied with Clinton.
In all of this, who is right and who is wrong? Who is in for the big "surprise" on Election Day: the majority of pollsters who are finding this race is unlikely to go Trump's way, the champions of Trump's message that the polls are missing something or perhaps both? How certain can we be that the polls are giving us an accurate picture of what is going on?
It is first important to note that neither blind faith in nor complete rejection of the polls is warranted. Being a skeptical and thoughtful consumer of polls is essential. When people note that more and more voters are cutting their landline phones and that more and more people are refusing to pick up phone calls from numbers they don't know, they are identifying problems that the polling industry has long struggled with and continue to try to adapt to.
Polling has tended to work fine for periods of time, until a major catastrophic miss causes the industry to re-examine and refine its methods. Literary Digest surveyed its readers and declared Alf Landon was going to be elected president in 1936, only to have to shutter its publication after that result proved wrong.
George Gallup then innovated in the arena of random sampling, but 12 years later, Gallup's research declared Dewey had defeated Truman. Gallup went back to the drawing board, fixed the mistakes in his methods and went on to build a polling empire that has thrived ever since.
Could we be on the edge of a Literary Digest moment, or a "Dewey Defeats Truman" embarrassment for the industry? Certainly. These errors have happened before, and in a moment where people are dramatically changing their information consumption and communication habits, it is possible that the polls could miss again.
But it is important to note that it is not a given that a polling "miss" due to methodological failure would necessarily benefit Trump. In fact, the very voters who are hardest to reach and to incorporate into polls are the same voters who are demographically inclined to vote for Clinton: young voters, Latino voters, those who live mobile lifestyles or who live in densely populated urban areas.
At the same time, those pointing to the 2012 polling misses, the Brexit vote or the 2016 presidential primaries as proof that the polls are "broken" often overstate their case. The truth is that polling has not actually had a "Dewey Defeats Truman" level moment driven by dramatically incorrect data. It is astonishing that, despite enormous declines in response rates, polls are still working fairly well when you actually look at the numbers.
Even in the 2016 Republican Primary, where Trump's victory shocked the world, it was not that the polls had been underestimating Trump's support, but rather that people were refusing to believe what the polls were telling them about Trump's rise.
In the 2012 election, the polls that had made Mitt Romney so confident that he was going to win were his own internal polls, based on models that failed to accurately estimate voter turnout. But the public polls, especially statewide polls, painted a fairly accurate picture of how the electoral college might go.
Polling averages had Barack Obama ahead by safe margins in states like Pennsylvania, and showed the race right within the margin of error in places like Florida, where Obama edged Romney very slightly. In critical Ohio, the RealClearPolitics final polling average showed Obama on track to win the state by 2.9 points; Obama wound up winning by 3.
There was much hand-wringing about the need for the polling industry to adapt after 2012, and certainly the high-profile polling miss by Gallup exacerbated the perception that the industry was in collapse. But by and large, Nate Silver was able to accurately predict the 2012 election because the statewide polls that were fed into his model were not so terrible after all.
Brexit is a more worrisome case study for the polling industry, but even there, the magnitude of the "polling miss" has been overstated dramatically. During the month of June, the polls were fairly evenly split between "Leave" and "Remain," and the final polling average had the race almost exactly tied.
The failure was much more about betting markets being off and commentators misinterpreting the evidence before them. Brexit surprised bankers and pundits, but the polls did not exactly suggest "Remain" was a slam-dunk.
The possibility remains that the polls are missing something badly, that a 5-point Clinton lead in the polling averages could be an illusion. But the examples of the 2012 election and Brexit are not sufficient to prove that polling is irreparably flawed and broken.
So then, with the polls tightening in the final days, the question remains: How likely is it that we will be surprised?
At the moment, Clinton still appears overwhelmingly likely to become the next president. The bulk of the polls in most of the key swing states she would need to achieve 270 electoral votes show her in a strong position, though one that has weakened as bad headlines about her email server have again dominated the news cycle.
States such as Colorado, once thought out of reach for Trump, have come back within the "margin of error," and critical prizes such as Florida and Ohio seem completely "gettable" for Trump based on the current polling averages. Trump needs a lot of things to go right for him to piece together the electoral votes to win.
He will need to win not only Florida and Ohio, but hold onto states such as North Carolina, red states that have slipped into swing-state territory. He will need to add states such as Nevada and Iowa to his column, and will need to pick up at least one state — such as Colorado — where he has not led in the polls even once in the last month.
But every poll has a different way of determining who is or is not a "likely voter," leading to what feels like an uncertain and wobbly race. When The New York Times gave four pollsters the exact same Florida voter survey data set a few weeks ago, each pollster came to a different conclusion about where the presidential race stood.
Some pollsters believe it is important to factor in a voter's known vote history before deciding if they are "likely" to vote. Others decide that it is more important to ask a voter about their own reported enthusiasm and likelihood to vote, despite the fact that studies show voters are notoriously bad at accurately reporting on their own vote propensity.
Some polls instead rely not on likely voters but a likely electorate; rather than a sample based on past vote history of individual respondents, pollsters look at the past makeup of an electorate to weight their survey results in such a way to reflect what the pollster thinks the current makeup of an electorate is likely to be.
In the case of the ABC/Washington Post poll that shows Trump surging in the last two weeks, they acknowledge that the change is not due to voters switching positions, but rather Trump voters being more likely to be considered "likely voters" in polls due to increased enthusiasm and self-reported likelihood to vote.
When Trump has a good news cycle (or Clinton has a bad one), Trump supporters are more likely to pick up the phone and say Make America Great Again. When Trump is having a bad news cycle, it is Clinton supporters who instead express more excitement. Whether this is actually having a material impact on who does and doesn't vote is still not clear.
It is possible that, in the end, for all the polling swings that have made this race feel like a roller coaster, the fundamentals of this race and the underlying attitudes of voters have been fairly stable.
One year ago, the RealClearPolitics polling average had Clinton ahead by around a 2-point margin. Today, RealClearPolitics shows Clinton ahead ... by around a 2-point margin.
Maybe the real surprise is that this is a race that is somewhat close, but slightly favors Clinton, and has been that way all along.
Kristen Soltis Anderson is a columnist for the Washington Examiner and author of "The Selfie Vote."