With less than a month to go until election day, both President Trump and former Vice President Biden are campaigning to see who will fill the Oval Office for the next four years. From the moment the two earned their party’s nomination, public opinion polling has shown Biden leading President Trump both nationally and in most swing states. The RealClearPolitics (RCP) average of national polls currently has Biden with a 9.2 point lead, similar national polling aggregators FiveThirtyEight and CNN show Biden with an 8.4 point lead and 11 point lead, respectively.

With that said, it’s important for voters to recall some of the disasters in public polling from the 2016 race when then-candidate Donald Trump pulled off perhaps the greatest electoral surprise in American history, winning the electoral college 304 to 227.

Looking at just the polling data alone in this year’s presidential election, you could swap out “Biden” with “Clinton” and you would see a very similar race to 2016. In fact, a recent CNN poll shows the former Vice President up 16 points on President Trump nationally. Interestingly enough, CNN released a similar poll at around the same time 2016 showing Hillary Clinton up on Donald Trump by 12 points—and we all know who won in 2016. The stunning collapse of the public polling industry in 2016, and the confidence in which they projected Clinton’s inevitable win, leave voters this year skeptical of the same prognosticators and pundits who failed to call the 2016 election correctly.

Voter registration data can inform us on existing trends in the swing states that will likely determine the election. It’s one thing to answer a poll on the phone, or online, and simply state your intentions to vote—it’s another thing to see how voters are actually behaving.

Among the most coveted swing states this year are Florida, North Carolina, and Pennsylvania. Each of these states report partisan registration data. A recent article from CBS News took a closer look at voter registration trends since 2016 and shows some interesting numbers. What does the voter registration data in these key swing states tell us?

In Florida, Republicans have outpaced Democrats in new voter registrations by about 183,000 new voters. Florida is the quintessential swing state. Diverse demographically, economically, culturally, and politically, small moves in the electorate can have huge ramifications for both state and federal elections. In 2018, Ron DeSantis won his race for governor by just 30,000 votes and Rick Scott unseated longtime incumbent Senator Bill Nelson by just 10,000 votes.

Pennsylvania also shows some interesting numbers of newly registered voters since 2016. Since 2016, when Donald Trump became the first GOP presidential candidate to win the Keystone State since Reagan in 1988, Pennsylvania Republicans have grown their ranks by 3.7 percent, while Pennsylvania Democrats have actually lost 1.5 percent of their share of the electorate. Just looking at voter registration numbers since June of 2020 shows a startling contrast: 135,619 new GOP voters registered in the state to the Democrats’ 57,985.

Of these three swing states, none show more promise for the president’s reelection hopes than North Carolina, where the number of registered Democrats has tumbled more than 6 percent since 2016, and GOP voter registrations have grown by almost 3.5 percent. In 2016, President Trump won North Carolina by 173,315 votes, or 3.6 percent. A shrinking pool of voters for Biden to pull from will undoubtedly make his task of improving on Clinton’s performance there in 2016 difficult.

Next, let’s look at two of the rust belt states that were reliably Democrat for decades and flipped to the GOP in 2016: Wisconsin and Michigan. Neither of these states register voters by party affiliation, so it’s hard to tell exactly how the major parties are faring. But both Michigan and Wisconsin make information on voter registration numbers by county available. Looking at changes in voter registration numbers at the county level and how each county performed in the last presidential election can give us a good picture of how each area of the state will vote in this year’s election.

Let’s start with Wisconsin—a state where then-candidate Donald Trump won 60 counties to Hillary Clinton’s 12. Since November 2016, the total number of voters in Wisconsin has dropped by more than 31,000. If we look only at the 60 counties Trump won in 2016, we see those counties actually gained more than 2,900 voters—a modest number in and of itself, but significant in that it bucked the statewide trend. When we turn our attention to the 12 counties Clinton won, we see a net loss of more than 36,000 voters. The bluest areas of the state are hemorrhaging voters.

In the electoral-vote-rich state of Michigan, Trump won 75 counties to Clinton’s 8. Michigan has seen its voter rolls grow by more than 400,000 voters since November 2016. Since then, the 75 counties that went for Trump reported an increase in over 245,000 new voters to the Clinton counties’ 155,000. Put another way, 61 percent of new voter registrations in Michigan are attributable to counties of the state where President Trump won, compared to just 38 percent from areas Clinton won.

Both Wisconsin and Michigan were decided by less than 1 percent. Wisconsin was decided by 22,748 votes, or 0.82 percent; Michigan was decided by a mere 10,704 votes, or 0.24 percent. Every vote matters, and with less voters in deep blue counties, we have to assume Biden will need to do far better than Clinton did in areas Trump won. Conversely, with more voters in areas Trump won, the president will be able to add to his column in these counties.

In each of the swing states we’ve discussed, Democrats still retain an advantage in overall voter registration. In North Carolina, that edge is around 400,000 voters; in Florida, the Democrat advantage is 183,000 voters; and in Pennsylvania, the Democrat edge is 717,000 voters. But the trend lines are clear: more new voters are opting to register as GOP or are registering in areas that lean heavily GOP.

It is also worth noting that the COVID-19 pandemic has not left the voter registration game unscathed. In fact, as colleges and universities remain shuttered in response to the virus, voter registration numbers in college towns have plummeted. This will lessen the impact deep blue college towns have on swing states as tens of thousands of out-of-state students who would otherwise be able to register to vote in the swing states mentioned earlier are forced to vote in their home state—which may or may not be competitive this year.

In closing, voters should be wary of reading too much into public opinion polls, and instead should look at other data. We’ve looked at how voter registration rolls in Florida, Pennsylvania, Michigan, North Carolina, and Wisconsin have changed. In states that allow partisan voter registration (Florida, Pennsylvania, and North Carolina), we have seen a clear increase in GOP voter registration numbers since the last time Donald Trump was on the ballot, and in some cases, we’ve seen the total number of Democrats actually decrease since 2016. In the two other states we looked at (Michigan and Wisconsin) that do not register voters by party affiliation, we looked at the county-level data and saw that areas where Donald Trump won in 2016 reported larger increases in voter registration numbers, while deep-blue areas where Clinton ran up the score either reported lesser numbers of new voters or even lost voters since 2016.

The last time Donald Trump was on the ballot, the odds looked to be stacked against him. Many national and swing state polling showed his path to 270 electoral votes to be practically nonexistent. Trump won in 2016 by pulling blue collar voters into his coalition and remaking American politics. While it is possible the pollsters have gotten their act together this year and are now telling us the actual state of the race, it is also just as likely they are making the same mistakes that undercounted the president’s base and failed to accurately predict turnout. The available voter registration data seems to be telling us that Trump’s base has actually grown since 2016, and this election will come down to who turns out to vote.