President Trump waves to the crowd on March 2. (Joshua Roberts/Reuters)

People are frustratingly fickle. While democracy would probably suffer if the only people allowed to vote were robots who were incapable of changing their opinions, it would certainly make political prognostication simpler. (Well, democracy might not suffer, as such, but I suspect we humans wouldn’t be thrilled with the country’s new laws.)

Consider a question asked in an NBC News-Wall Street Journal poll released over the weekend. According to that survey, 41 percent of respondents indicated that they will definitely or probably vote for President Trump’s reelection next year. By contrast, 48 percent say they definitely or probably will vote against him. In other words, nearly half the country says at this point that they will vote for the Democratic candidate on the 2020 ballot — about the percentage that Hillary Clinton earned three years ago.

That’s the worst margin on this question that any incumbent has seen at the equivalent point in the last four presidential bids for reelection. Bill Clinton was underwater on the question, but only 42 percent of Americans said they’d vote against him. George W. Bush, still enjoying a boost from his handling of the Sept. 11 attacks in early 2003, had broad support for his reelection. In 2011, Barack Obama was ahead on this metric to the same extent that Trump is now behind.

(Philip Bump/The Washington Post)

But here we get to the robot issue: People change their minds and circumstances change. Clinton ended up winning, of course, earning 11 percentage points more in the actual vote than the 1995 poll indicated. For Bush, who had a broad lead on this metric, things narrowed dramatically.

(Philip Bump/The Washington Post)

Really, the past three elections have shown the three possible shifts one might expect to see in a predictive poll: In one case, the incumbent overperformed, in one he underperformed and in one both numbers increased about the same amount.

Which is a long way of saying that those past numbers aren’t really predictive. That said, it’s worth highlighting some reasons that the numbers changed. Clinton had a strong economy and a weak opponent. Bush had mounting opposition to his war in Iraq (which had only just started when the poll was conducted). Obama? We’ll get to him in a second.

First, let’s look at another number that’s not good news for Trump: Independents are 18 percentage points more likely to say they’ll vote against Trump than vote for him. Democrats plan to oppose him (no shock) and Republicans to vote for him (also no shock). But those independent numbers seem ... problematic.

(Philip Bump/The Washington Post)

Especially when compared to where Obama stood. He was at plus-7 overall, you’ll remember, and was plus-3 with independents. Exit polling suggests that he ended up losing independents by about five points, even as he won the election by four points. But notice that his standing with independents dropped. If Obama won the popular vote by four points while losing independents by five, it doesn’t suggest good news if Trump’s support from independents remains soft.

(Philip Bump/The Washington Post)

We didn’t highlight the change in those independent numbers above, though, instead focusing on the change in opposition from Republicans. In 2012, Obama’s position also softened as Republicans “came home” — that is, they ended up voting for the Republican. That year (when the poll question asked if people would probably vote for the incumbent, not differentiating out those who “definitely” would), Obama had the support of 84 percent of his party in the NBC-WSJ poll and had 78 percent of Republicans opposing him.

Now? Feelings are more cemented: 86 percent of Republicans already plan to support Trump and 90 percent of Democrats plan to oppose them. Those numbers could shift, but it doesn’t leave Trump much room to grow. What’s more, two-thirds of Democrats say they will definitely vote against him while about 6-in-10 Republicans say they’ll definitely vote for Trump.

This is a lot of analysis for an article that also includes the accurate line that past results haven’t been predictive. But it does suggest that Trump’s standing with independents is in problematic territory.

Happily for the president, those independents are overwhelmingly not robots who won’t change their minds.