John Podhoretz explains in the latest issue of Commentary magazine why he recommends that we “throw out the polls.”

I write on the morning after the British general election, in which the ruling Conservative Party won an absolutely staggering victory that has entirely redefined Britain’s political future. The most important third party over the past three decades (the Liberal Democrats) is likely dead and buried. Its successor as the key third-party force, the Scottish Nationalist Party, has taken a deadly bite out of the hide of the Labour Party. The right-wing populist UKIP party gained only a single seat in Parliament but received three times as many votes as the Scottish Nationalists, who won 58. With that kind of following, UKIP may be in a position in future elections to do to the Conservative Party what the Scots did to Labour.

And the key point is this: Nobody saw any of this coming. Nobody.

Commentators and pollsters alike were writing pre-obituaries on the morning of the vote about the cautionary lessons to be drawn from the seeming failure of Tory leader David Cameron’s “modernization” program. And why not? Every piece of social-science data suggested a tight race between the Tories and Labour—and all of it was wrong. But Cameron’s party ended up with 105 more seats than Ed Milliband’s Labour and a majority in the House of Commons for the first time in 23 years.

These results came six weeks after the stunning outcome of the Israeli general election, in which Benjamin Netanyahu’s Likud party rose from 18 to 30 seats in the Knesset. Again, no one had even dreamed of such a thing only 24 hours before the votes were counted. Polling—and there was a lot of it—had Likud and the Zionist Union neck-and-neck.

Similarly, in the American midterm elections in 2014, in state after state, polling had Senate elections too close to call in races that were won by five, six, eight, even 11 points when the votes were all in.

There is a commonality to these failures across three very different political systems with very different political realities. First, they suggest that polling as it has been conducted for the past 70 years is now irredeemably broken—that it has not figured out how to move forward following the death of the landline telephone as the primary means of communication.

Second, after a refreshing moment in 2008 when the polling in the United States proved remarkably accurate, we appear to be reverting to a traditional problem: The systematic underrepresentation of right-of-center voters in these surveys.