Thursday was a National Day of Prayer, per a proclamation from the White House. President Trump hosted religious leaders in the Rose Garden and offered some remarks on the subject at hand.

It was a continuation of a tradition that dates to 1952, the last year of the administration of Harry S. Truman. It’s continued since, even as the religious identity of Americans has shifted. Pew Research Center data from 2015 showed that even as the number of American adults who described themselves as religiously affiliated declined from 2007 to 2014, the percent who were both religious and who prayed daily held steady. A bit more than half the country overall reported praying daily in 2014, down slightly from seven years earlier.

In light of the event, we were curious about how America’s religiosity had evolved over the 60-plus years since the first National Day of Prayer. Recognizing the distinction between church attendance and prayer, we looked at data from the Association of Religion Data Archives to assess how church membership has evolved over the event’s history — with an eye toward geographic changes.

That data suggested an interesting trend: Big swaths of the middle of the country show a higher density of the population attended church than 60 years prior, with declines in that density in the northeast and Midwest.

This runs counter to our understanding of how church attendance has evolved. We tend to believe, thanks to research like Pew’s, that church attendance has waned. (From 2007 to 2014, the percentage of Americans who reported attending services at least monthly dropped from 54 to 50.) The map above shows many places where there was an increase.

In fact, the number of counties where church attendance increased since 1952 as a function of population is distributed fairly evenly, with slightly more counties seeing increased density in church attendance than decreases.

How can that be the case when religion is declining in the U.S.? For the same reason that maps of the 2016 election results look like a landslide win by Trump: We’re considering counties, not population.

If we look at the shift since 1952 relative to the 2010 population density in each county, there’s a clear pattern. Lower-population counties, generally more rural ones, saw much larger increases in the density of the churchgoing population than did larger counties.

This is a bit confusing, so an example might help. Imagine a county that was home to 1,000 people in 1952 and 2,000 in 2010. In 1952, 100 people were members of congregations, a rate of 100 people for every 1,000 residents. In 2010, 400 of the 2,000 residents went to church regularly, a rate of 200 people for every 1,000 residents. That’s a big increase in the density of the church-going population — but it’s an increase of only 100 more church attendees.

Truman was born in Barton County, Mo. It saw a substantial increase in the density of the churchgoing population from 1952 to 2010, according to the ARDA data. Trump was born in Queens County, N.Y. The density of churchgoers there has declined.

Seems fitting, given the national trends.