America is often described as a place of great divides — between red and blue, big cities and rural towns, the coasts and the heartland. But our social lives are shaped by a much stronger force that ignores many of these lines: distance.
In the millions of ties on Facebook that connect relatives, co-workers, classmates and friends, Americans are far more likely to know people nearby than in distant communities that share their politics or mirror their demographics. The dominant picture in data analyzed by economists at Facebook, Harvard, Princeton and New York University is not that like-minded places are linked; rather, people in counties close to one another are.
Even in the age of the internet, distance matters immensely in determining whom — and, and as a result, what — we know.
Coastal cities like New York, Washington, San Francisco, Boston and Los Angeles do exhibit close ties to one another, showing that people in counties with similar incomes, education levels and voting patterns are more likely to be linked. But nationwide, the effect of such similarity is small. And the pull of regionalism is strong even for major cities. Brooklynites are still more likely to know someone on Facebook near Albany or Binghamton than in the Bay Area.
9 Nantucket County, Mass.
10 Highlands County, Fla.
11 Okeechobee County, Fla.
12 Miami-Dade County, Fla.
This map shows an index of connectedness, created using friendship links between pairs of anonymous Facebook users from a snapshot of the platform in April 2016. The researchers aggregated the links at the county level, so neither the Times nor other academics working with the data can identify individual users within it, or how many Facebook users live in each county. Because counties with more people invariably have links to more places, this map rescales the index to account for differences in population.
The darker the color, the greater the relative likelihood that any two people living in two different counties are connected on Facebook. Counties that are broadly tied to more — and more distant — places will color in more of the map. Counties where nearly all ties are very local will look more isolated.
The networks that emerge reveal a distinct social fingerprint for each county, influenced by past migration patterns, geographical features and quirks of the local economy — whether, for example, the county is home to a military base, a resort hub or a booming oil industry.
“But distance is the thing that shows up everywhere,” said Johannes Stroebel, a professor at N.Y.U.’s Stern School of Business and a co-author of the research, along with Michael Bailey at Facebook, Rachel Cao at Harvard, Theresa Kuchler at N.Y.U. and Arlene Wong at Princeton. “Distance matters,” Mr. Stroebel said, “in explaining every single county’s connectedness.”
Even in Washington, D.C., nearly half of friendship links extend to people who live within 100 miles. Nationwide, in the average county, 63 percent of friendship links are that close, probably reflecting that many people on Facebook know one another through real-world sites like grade schools, colleges and offices. Other research shows that these sites tend to be close to home: The typical American lives just 18 miles from his or her mother. The typical student enrolls in college less than 15 miles from home.
Proximity works alongside every other pattern shown below.
State lines are powerful boundaries in binding nearby places. For many counties, as the maps illustrate, the likelihood of friendship drops off sharply at state borders. And counties within a given state tend to be strongly connected to one another. This is particularly striking in Michigan, where counties near the Indiana and Ohio state line are more closely tied to Michigan’s Upper Peninsula than to out-of-state counties closer by.
History is another significant force in shaping these networks. Decades-old migration patterns can help explain why some distant counties are disproportionately connected today. Northern cities like Chicago and Milwaukee still retain close ties to Southern counties along the Mississippi River, where African-American workers who were part of the Great Migration starting a century ago left communities for industrial jobs in the North.
Kern County, Calif., holds ties today to counties in Oklahoma, where the Dust Bowl in the 1930s led to another significant migration (oil-producing Kern County today is also closely tied to parts of northwest North Dakota that are experiencing an oil boom).
In other parts of the country, physical geography forms a kind of social barrier. Friendship links from Belmont County, Ohio, extend east but don’t cross the Appalachian Mountains in Pennsylvania or the Blue Ridge range in Virginia. In Scott County, Ark., friendships do cross state lines, into Oklahoma, Louisiana and Texas. But they don’t cross the Mississippi River to the east. In Nassau County, N.Y., the likelihood of friendship links declines steeply off Long Island.
Other county outliers in the data can be explained by distinct roles some communities serve: Onslow County, N.C., which is connected to much of the country, is home to Marine Corps Base Camp Lejeune.
Charlotte County, Fla., a retirement community destination, is disproportionately connected to counties across the northern Midwest and Northeast. Washington, D.C., a magnet for college graduates from across the country, is closely tied to counties that are home to major state universities in Florida, Alabama, Wisconsin, Iowa and Michigan.
The power of distance underlying these other patterns can be seen another way: If we were to divide the United States into two regions, merging counties that are most closely connected to one another, we would get a very simple map. It would not show the coasts versus the heartland, or red America versus blue America.
It would show, simply, all of the continental U.S. and Alaska in one region, and far-off Hawaii in the other. Divide the country further, and cohesive regions become clear at different scales. Northern Florida merges with southern Georgia. Texas and California splinter. Divide the country into 50 regions, and you get something that looks like how we might redraw our state borders to reflect the social worlds people in America inhabit today.
This is what happens when you split the country into two parts.
Hawaii is its own region.
This is what happens when you split the country into three parts.
The West and the northern Plains become a region.
… into four parts.
Then the northern Plains becomes its own region.
… into 10 parts.
Texas starts to approximate its current form.
… into 20 parts.
States’ shapes start coming into view.
… into 50 parts.
If state lines were redrawn this way, California would span four states.
… into 100 parts.
Now we can answer questions like: What is downstate Illinois or upstate New York?
… into 150 parts.
Or West Texas?
… into 200 parts.
Metropolitan areas within states are now clear.
… into 435 parts
Now we’ve redrawn every congressional district in America (albeit not with equal population).
These networks are important in part because of other patterns that are correlated with them. Counties with more dispersed networks — where a smaller share of Facebook friends are located nearby, or among the nearest 50 million people — are on average richer, more educated and have longer life expectancies. Places that are more closely connected to one another also have more migration, trade and patent citations between them.
Counties that are more geographically isolated in the index are more likely to have lower labor force participation and economic mobility, and they have higher rates of teenage births. Some of the most economically distressed parts of the country appear to be the most disconnected: Among the 10 U.S. counties with the highest share of friends within 50 miles, six are in Kentucky.
Where most friends live within 50 miles
Close-knit communities can have their own benefits, like enabling neighbors to rely on one another for economic and social support. But previous research suggests that “weak ties” to people we know less well can be particularly valuable for bringing us information we don’t already have. So people in communities that are more broadly connected may be more likely to hear about a wider range of business or educational opportunities.
The patterns in this Facebook data don’t necessarily mean that limited social networks cause worse economic and health outcomes, or that wide-ranging networks produce better ones. But other researchers say this data will make it possible in future studies to untangle why they’re related.
“This gives us the first way to systematically look at some of those relationships,” said Mark Granovetter, a sociologist at Stanford who has written influential papers on the value of social networks. “They have just scratched the surface here.”
Facebook links used in the research are between users active on the platform within 30 days of the April 2016 snapshot. Users in the study were assigned a county based on the location in their Facebook profile, their activity on the site and their regular login sources. Some independent cities in Virginia are missing from the data.