It’s hard to know with certainty what voters will do before they do it.

In 2016, for example, we understood from polling that both Donald Trump and Hillary Clinton were viewed negatively by the electorate. We also knew that, comparing the two major-party candidates on their personal qualities, more voters said that they had confidence in Clinton than in Trump. Over the course of the campaign, Clinton’s lead rose and receded, with a large chunk of the electorate withholding an opinion.

For those observing the pattern, the result wasn’t clear. Would the race end with a big Clinton lead or a small one? Would those indifferent voters vote or stay home? If they voted, who would they support?

Once the vote results were tallied, we had our answer. The 2016 race narrowed in the last days of the campaign, with undecided voters shifting to support Trump. Some of the voters who didn’t like either candidate probably stayed home, but those who voted preferred Trump by a 17-point margin over Clinton.

From the outset of the 2020 contest, it has seemed as though Trump was hoping to replicate this strategy against former vice president Joe Biden. If he could make Biden unpopular, push his own supporters to the polls and win over those who made up their minds late, maybe he could win two contests in a row.

So far, however, that’s not how things have worked out. Biden is more favorably viewed than Trump by a healthy margin and much more so than Clinton was four years ago. His lead over Trump has been much more stable; at no point has there been any sort of tightening similar to the wavelike pattern in 2016. In states that polling shows are going to be close, Biden maintains leads.

The question for Trump is whether he can shift that trend over the next two weeks. Yes, the polls were off in key states in 2016, but they weren’t off by so much that one can dismiss Biden’s current leads in a number of states as shaky. Nationally, Trump trails by double digits, according to FiveThirtyEight’s average of the polls, the sort of margin that essentially precludes him being able to win the electoral college.

What observers will want to track, though, is how those numbers change over the next two weeks — or how they don’t. To that end, we’ve created the tool below, showing the state of the race in the last 28 days in both 2016 and 2020.

Support. This column shows the polling average for the Democrat (blue line) and Republican (red) in 2016 (dotted line) and 2020 (solid line) over the last 28 days of the race. The final results in 2016 for Trump and Clinton are shown by a circle at the right side of the column.

Margin. The current (solid) and 2016 (dotted) margins by day over the last 28 days. When the solid line is lower, it indicates moments at which Biden’s lead was larger than Clinton’s. The circle at right indicates the final margin.

Undecideds/third-party support. The percentage of respondents who either indicated that they were undecided or planned to vote third-party. The final percentage voting third-party is indicated by the circle at right.

Before we get into what it suggests, take a look at the graphic.

Your browser cannot display this graph.

We can start with the undecided numbers. At this point in 2016 (that is, as of this writing), there were far more undecided voters and voters planning to vote third-party. That gave Trump much more space to pad his support, particularly given that he was running as the outsider, a role that often benefits from the support of people making up their minds late.

Biden also has a wide margin nationally and in various states, such as Michigan and Nevada. These are the places where Trump needs to eat into the Democrat’s lead if he’s to have a shot. (Two of the Trump campaign’s three victory scenarios require at least one of these states.)

In other states, Biden’s leads are closer to where Clinton was four years ago, including Pennsylvania and Wisconsin. But, again, those leads began to narrow a bit as Election Day approached, and, in each state, there were more undecided voters from which Trump could draw support.

That’s the problem for Trump in many of these contests. It’s not just that he is trailing; it’s that Biden’s support is over 50 percent, a mark that Clinton didn’t hit often in the last month of the campaign in swing states. More people have made up their minds, and, in many states, the majority back Biden.

Look, for example, at the data in Michigan. In the last two weeks, the number of undecideds dropped as Trump gained support. Clinton added basically nothing. The trend was all Trump, and he narrowly won the state. This year, he can’t catch Biden without stealing support from Biden, something that, so far, he hasn’t had much luck doing.

Another way of looking at the state of the race is to compare where the race is now with where it was at the same point in 2016 — and how the race had changed over the preceding week.

Four years ago, the shift to Trump was already underway in most states. This year, it isn’t.

Just because the race hasn’t changed yet doesn’t mean it won’t. Biden’s support could erode to Trump’s benefit. The incumbent could benefit from the remaining pool of undecided voters. His people could turn out to vote while Biden’s stay home. We won’t know until the voters vote.

As it stands, though, the pattern is not one that suggests an easy reelection for Trump.