President Trump currently trails his presumptive Democratic opponent, former vice president Joe Biden, in nearly every national public opinion poll. But a closer look at the data reveals he may have a secret stash of voters that makes the race closer than it currently appears.

That stash might come from a curious polling anomaly. Every poll in the RealClearPolitics average that asks both a job approval and a Trump vs. Biden ballot question finds that more people approve of Trump’s job performance than say they will vote for him. The gap between job approval and vote share ranges between 2 points in the Monmouth and CNN polls and 7 points in the Fox News and CNBC polls, but every poll has a similar finding.

This gap did not exist in President Barack Obama’s 2012 reelection campaign. Polls from April 2012 had no consistent pattern. Four polls found Obama’s share of the vote in trial heats against Republican nominee Mitt Romney was higher than his job approval rating, while three showed the opposite (one had them the same). All of the differences were no more than three points, which means the differences were within the poll’s margin of error. The result was that Obama’s average share of the vote largely tracked his job-approval rating.

This symmetry between a president’s share of the vote and his job-approval rating is what normally happens when presidents run for reelection. Election analyst Dan Guild recently examined the relationship between job approval and presidential vote share for Sabato’s Crystal Ball. He found that a president’s final share of the popular vote almost precisely tracked his final job-approval rating at both the national and state levels. The only exceptions were in 1992 and 1996, when independent Ross Perot ran strong campaigns, winning 19 percent in 1992 and 8 percent in 1996. Those exceptions went away, however, when Guild stripped out votes cast for third-party candidates such as Perot; the president’s share of the two-party vote, he found, largely tracked the final job-approval ratings.

Biden’s lead over Trump would shrink considerably if this pattern holds true again. Biden leads Trump by nearly six points, 48.3 to 42.4 percent, in the most recent RCP average. Trump’s approval rating in the RCP average was 46.0 percent on Wednesday morning. If Trump’s true vote share approximates that, he only trails Biden by about 2 points. If that happens on Election Day, Trump could once again win the electoral college while losing the popular vote.

There’s some evidence to suggest Trump’s uniquely polarizing personality will break the pattern, which Guild’s analysis shows has been going on for 48 years. Many of the polls in the RCP average also measure Trump’s personal favorability ratings. Trump polls worse on that score than he does on his job approval, garnering only 43.4 percent in the RCP average. That’s much closer to his current share of the vote against Biden, and this relationship holds for most of the individual polls as well. Two polls currently find Trump receiving one or two points more against Biden than he gets in personal favorability, even as they show him doing three or four points worse against Biden than his job-approval rating. The Quinnipiac poll shows Trump’s vote share against Biden is the same as his personal favorability (41 percent) but is four points lower than his job approval. It could be that people will vote this time against a president whom they dislike, even if they grudgingly think he’s doing a good job.

Other data, however, point in the opposite direction. Most of the polls show very little difference between job approval and vote share among partisans. The Quinnipiac, CNN and Fox News polls, for example, show that Republicans and Democrats are both about as likely to say they will vote for Trump as they are in saying they approve of his job performance. Each of those polls, however, shows a big difference on those measures among independent voters. In each case, a much higher percentage of independents say they approve of Trump’s job than say they will vote for him. The Trump campaign can reasonably expect that independent voters who approve of his job performance will eventually come around to voting to reelect the president.

It’s been a clear rule for the past eight elections that a president’s vote share is about the same as his job-approval rating on election day. Trump has broken most of the old rules during his short time in politics. He’d better hope this rule holds true if he’s going to win a second term.

Read more: