Of all the lessons the mainstream media should have learned over the past four years, one of them is inexcusably obvious: We’re bad at predictions.

Who can forget poll analyses, the infographic needles, the endless columns and pundit panels that confidently projected that Hillary Clinton would be the next president?

In fact, it was that very certainty that drove much of the damaging and excessive coverage of her email practices. (Because she was all but certainly the next president, went the flawed thinking, we should get started now on showing we’ll be tough on her.)

Washington Post columnist Dana Milbank was so sure, in late 2015, that Donald Trump would fail to get the Republican nomination that he promised to eat his printed column if he was wrong. (An honorable man, he followed through.)

But my colleague was far from alone. And the following year, most media people were stunned by Trump’s winning the presidency.

And yet, we roll on today, with very little having changed.

The Iowa caucuses, according to many a commentator, surely would come down to a fight between Joe Biden and Bernie Sanders. As we know now, it didn’t work out that way: Biden finished a lackluster fourth and the photo finish with Sanders involved Pete Buttigieg.

Oh, and surely, we heard many times, voter turnout at the caucuses would be high, signaling the voters’ intense desire to replace President Trump with a Democrat. Well, that didn’t happen either. Turnout was flat.

Journalists — reporters and opinionators alike — just aren’t very good at this. Yet they go on doing it, in part because there’s so little accountability in political journalism.

“There’s no penalty for being wrong. You can just blithely go on making another evidence-free prediction with no consequences,” Columbia University journalism professor William Grueskin said.

He added that it wouldn’t work that way in, say, business reporting: If you predicted that “XYZ Corp will lose $100 million” and then XYZ reported a $250 million profit, “much of your audience would have a hard time taking you seriously again.”

Part of the problem is our endless desire to frame politics as a horse race. Rather than focus on what’s most important to the public interest, political coverage focuses far too heavily on who’s up and who’s down — an elusive metric that requires data such as polls and fundraising numbers.

But these things are far from a reliable basis for prediction. What’s worse, many journalists don’t understand basic notions underlying polling, as data-oriented journalist Nate Silver has emphasized over the years: “Media understanding about probability, margin of error and uncertainty is very poor.”

If a candidate has an 80 percent chance of winning a race, no one should be surprised if he or she loses. The use of predictive phrases such as “all but inevitable” should be confined to instances where the chances of a candidate’s election are more like 98 percent, he says. (Silver has become more careful since he infamously stated during the 2016 presidential primary season that Trump’s candidacy shouldn’t be taken seriously. When proved wrong, he followed up by publishing a mea culpa about what went wrong in his thinking, and he was more accurate than most about Trump’s chances in the general election.)

Silver’s organization, FiveThirtyEight, remains solidly in the horse-race business but with a greater effort now to explain probability and to resist the urge to walk out onto flimsy limbs.

As for me, I’m officially done with making predictions, after flubbing one in mid-2017. New York magazine asked me, along with 12 others — journalists, academics and political types — to predict whether Trump would make it through his first term.

Some respondents posited that he’d be gone by now, for various reasons; others were convinced that he would survive and thrive. I managed to get one thing right: that Trump would make it through his first term, but I predicted that he would choose not to run again — and that Sen. Kirsten Gillibrand (D-N.Y.) would become the first female president. (I’ll note in my own defense that this was before she led the charge that brought about Al Franken’s resignation asthe junior senator from Minnesota.)

Since then, I refuse to predict — not in print, not in interviews, and not even in informal conversations with friends and family members.

In a recent MSNBC appearance, the author Anand Giridharadas suggested that, given their generally lousy records, media people should quit the prognosticating game. He put it simply and correctly: “I don’t think it is our job as journalists to predict.”

As we head deeper into the presidential primary season, with all of its temptations, it’s high time to heed that advice.

For more by Margaret Sullivan visit wapo.st/sullivan