Those who are paying close attention to the day-to-day developments in presidential polling might want to peruse historian David Greenberg’s essay in the latest issue of The Atlantic.

In November 1975, one year before the obscure Georgia governor Jimmy Carter was elected president, the field of Democratic presidential aspirants was in chaos. According to the polls, voters’ top choices were Senator Ted Kennedy of Massachusetts (23 percent), Governor George Wallace of Alabama (19 percent), and former Vice President Hubert Humphrey (17 percent). Unfortunately, only one of these men—a widely reviled racist—was actually running. To be sure, there was a grass-roots favorite expected to vault into contention by winning the Iowa caucuses, but it wasn’t Carter. It was Senator Birch Bayh of Indiana. Carter was netting low single digits. Newsweek explained that he could become viable “only in the long-odds event that [he] can stop George Wallace” and get the southern vote.

Four years later, Carter was, of course, president. And the late-1979 polling data strongly suggested that he would be dethroned—by Ted Kennedy. The great liberal hope for Democrats despairing of Carter’s incompetence, Kennedy had been scoring 60 percent in matchups against the incumbent earlier in the season. In the late fall he was still favored by Democratic majorities. One New York Times survey found black voters choosing Kennedy over Carter 53 to 15 percent, conservative Democrats favoring him 58 to 22 percent, and even southerners backing him 44 to 29 percent. But in the end, Kennedy triumphed in only 10 states, mostly in the Northeast.

In 1987, after the Democratic front-runner, Gary Hart, withdrew from the race following questions about his private life, the new leader, with 25 percent in the polls, was a candidate with no real prospect of winning: the Reverend Jesse Jackson. Jackson’s chief adversary was the Illinois senator and unreconstructed liberal Paul Simon, who was surging in Iowa. The party’s eventual nominee, Michael Dukakis, was in the fight, especially in New Hampshire, but The New York Times noted, “Recent surveys show him to be increasingly vulnerable in the state.”

Even further from the front of the pack one year out was Bill Clinton in 1991. While he was hovering around 6 percent, Democratic voters were overwhelmingly telling pollsters that they preferred New York Governor Mario Cuomo, who never entered the race. Among the announced candidates, the leader was California’s Jerry Brown, who was at that time known for his flirtations with Zen Buddhism, his “Governor Moonbeam” nickname, and the French filmmaker turned aide-de-camp who followed him around in a black beret. Brown was, in other words, a national joke.

Why have past polls proven to be such poor predictors of presidential preference?

The first thing to keep in mind is that a year before the general election, most voters aren’t paying attention yet. Campaign reporters and political junkies—and probably anyone reading this article—are apt to forget this, because we talk about the race incessantly ourselves. But only about 10 to 20 percent of voters are tracking the campaign closely. Normal people tend to tune out the arcane, minute developments that the Twitterati are quick to label game changers. Believe it or not, they have better things to do.

Their indifference may be justified. A California voter with a modest interest in public policy has no good reason to figure out whether she prefers John Kasich or Rand Paul or Carly Fiorina on the Islamic State or abortion or entitlements, because by the time she votes next June, there’s a good chance none of them will be in the race. She will begin studying the candidates carefully only in the days just before her state’s primary.

This inattention means that early poll numbers are based on shallow preferences. “The media don’t always report the numbers that say ‘not sure’ or ‘don’t know enough,’?” says David Karol, a political scientist who has studied the nomination process. Many people who are actually undecided, he adds, will cough up a name when a poll-taker calls and prompts them. Those responses just don’t tell us much.

It’s interesting information for both fans and foes of the current Trump phenomenon.