The pollster who nailed the Pennsylvania result in the 2016 presidential race almost exactly has a new poll that shows President Donald Trump narrowly leading Democratic nominee Joe Biden.

Based in Atlanta, Georgia, the Trafalgar Group poll 48.4% of its surveyed voter backing or leaning toward Trump with 47.6% backing or leaning Biden.

Almost no other pollster has Trump leading in Pennsylvania.

Trafalgar surveyed 1,076 likely voters and says its poll has a margin of error of plus or minus 2.91 percentage points. The margin of error means each candidate’s polling number could be that much higher or lower.

In a 2016 poll that ended three days before the election, Trafalgar had Trump winning by about 2 points. He won by about about three-quarters of a point. Pollsters who regularly poll Pennsylvania voters had Hillary Clinton ahead in the days leading up to Election Day.

Back then, Trafalgar founder Robert Cahaly said he detected a bias in primary election polling. The bias showed people were more likely to say they support Trump if they only had to push a button to signal their support rather than talk to a live interviewer.

Recently, a writer for 538, the poll-tracking website, took on this bias and said he found no evidence that it’s happening this year, but there’s no doubt Trafalgar was right in Pennsylvania in 2016 when others weren’t.

538’s average of recent polling shows Biden ahead. Perhaps more significantly, multiple polls have Biden with at least 50% support, which wasn’t true for Clinton in 2016.

In other interesting findings in the company’s poll:

— Biden leads among women voters, 52.5% to 41.9%, but Trump leads among men, 52.5% to 44.7%. The numbers include leaners.

— In age groups, Biden leads solidly among 25-34-year-olds (50.5%-44.5%) and narrowly among senior citizens (50.9%-47.3%). Trump leads solidly among 18-24-year-olds (51%-43.5%) and 35-44-year-olds (51.8%-41.8%) and narrowly among 45-64-year-olds (48.5%-48%). That Trump advantage among 18-24-year-olds stands out as a real surprise. It could be a poll flaw or insight. We’ll see.