Last night, as returns were coming in, we experienced a familiar phenomenon: People seizing upon early election returns as being significant. People should not do that.

Cochran supporters watch returns come in. (AP Photo/Rogelio V. Solis)

There are a few things that you should and can pay attention to, which we'll get to in a second. But first, let's look at why early returns are useless.

Shortly after 9 p.m. on Tuesday, I took a screenshot of the results for New Jersey. In several races, early returns were just coming in. Here is how two of those races' final results compare with the early returns.

In one case, the results were completely wrong. In the other, pretty on the money. What gives?

The entire point is that you can't really know. There are a number of things that might come into play that affect how returns will evolve.

Perhaps the most important consideration, if you want to track how a race is going, is where the returns are coming from. Returns come in as precincts turn over clumps of votes to whoever's doing the counting. That means that certain areas tend to add a bunch of votes at a time.

Look at the map we created earlier today of the results in Mississippi. This shows the vote margin by county for Chris McDaniel in that race.

If you'd only gotten a bunch of precincts from Hinds County, the light-colored blotch to the west of Jackson, the results would be lopsided for Cochran. If you'd gotten them for Jones County, the blotch near Hattiesburg, the opposite would be the case.

Knowing what geographies are important is not simple, by the way. Only people who'd done a lot of homework or who knew the state well would know the likely importance of Jones or Hinds Counties. What's more, not all of the places that provide returns tell you where they're coming from as they come in.

What you can do, though, is watch how the vote count changes instead of watching the totals. As results come in, the speed with which the lead changes will necessarily decline (after all, adding 100 votes to 1,000 total counted is different than adding it to 100,000). Last night, in that Republican primary, Bell pulled slightly ahead of the pack, and, as more returns came in, his lead grew a bit and then held steady. If anyone were to catch him, they would need to start narrowing the gap. Only very rarely does a massive cluster of votes from a heavily partisan area come in to turn the tide of a race. That's what trailing candidates hope for, but it's not what you should expect.

Here's an example of why this is informative, in a series of tweets from Cox Radio reporter Jamie Dupree. They run bottom-to-top.

The trend? Nothing's changing. Ratcliffe is in good shape. The odds that things would change dramatically after staying static as a third of the vote came in is unlikely. And sure enough: Ratcliffe won, 53 to 47.

While we have you, another note: when a lead changes, there's a moment at which the candidates are nearly tied (because one candidate passes the other). That doesn't mean they really are tied, however, and is just another reason to watch the trend -- who's going up, who's going down. If 60 percent of the vote is in and Candidate A leads 45 to 35, and then 62 percent is in and it's 43 to 37, and then 65 percent is in and it's 40 to 40, the story is not, "Oh, this is a close race!" The story is, "Well, this isn't going well for Candidate A."

How, you may ask, does the Associated Press (or whoever) call a race shortly after polls close? There are a few ways. In presidential races there are exit polls that provide a guide. The AP also has people that know the geography of these places, and tracks where votes are coming from.

And sometimes, a race is simply a blow-out between an established candidate and people who didn't spend any money and didn't really campaign. If one percent of the vote is in between Hillary Clinton and 24-year-old Jim Taylor of Hoboken, and Hillary is leading by 95 percent, go ahead and assume she's going to win. But still watch the trend.