Race and education make a bigger difference in who you vote for than ever before

If you’d told me before the 2016 election that the Democratic candidate was going to finally win Orange County, California (the most legendary Republican stronghold of all), as well as Fort Bend County, Texas, and Cobb County, Georgia (the suburbs that gave us Tom DeLay and Newt Gingrich, respectively), it would have been safe to assume that would be part of a crushing Democratic victory, something on the order of LBJ vs. Goldwater.

On the other hand, if you’d told me that the Republican candidate was going to win places like Kenosha County, Wisconsin, Trumbull County, Ohio, or Monroe County, Michigan (bastions of organized labor that kept the Democratic faith over the decades, even during their 1980s low-water mark), that would have sounded like a catastrophic Dem wipeout, probably wore than Walter Mondale’s benchmark of futility.

But if you’d told me that both things would happen in the same election, if I hadn’t slowly and cautiously backed out of the room, I might have assumed you were talking about a distant future election, one that exaggerated trends we were only starting to see. And yet, that’s exactly what happened in the 2016 election. Previous decades of county-level results had correlated fairly strongly with how racially diverse they were, and in the last decade or so, a relationship with levels of education also started to develop. In 2016, though, those correlations both became much stronger, to the extent that the Dems started winning in almost all places that are diverse and well-educated—even Sun Belt suburbs with a long Republican tradition—and the GOP started winning in almost all places that are almost all white and where there aren’t a lot of college grads, even the former bulwarks of organized labor.

That’s an idea that I’ve been exploring in detail since the county-level election results for 2016 became available, and I’d like to share a new framework for thinking about what happened. Usually we think of political trends at the state level: a good idea, considering the importance of the Electoral College, but states aren’t very uniform units, given that the largest state (California) is around 80 times more populous than the smallest ones, and that most states contain a variety of urban and rural, and white and diverse, places. Sometimes we think in terms of congressional districts instead, which tend to be demographically uniform and all around the same population—but their boundaries change at least every decade, so that’s not helpful for viewing long-term trends.

Instead, what I’m proposing is that we create a number of buckets (25, in my example) that include all counties with a certain demographic makeup. What I’ve done is divide the counties into five quintiles of diversity (in terms of the percentage of non-Hispanic white population), and five quintiles for education (in terms of the 25+ population with a bachelor’s degree or higher). By overlapping these two sets of quintiles, you can create a matrix of 25 “states” that aren’t contiguous, but have similar populations (usually 10 to 12 million, but ranging from 6 to 24 million at the margins) and are demographically pretty uniform.

For instance, you can look at the bucket of counties with a very high level of diversity and average education (which tend to be our biggest cities), or a low level of diversity and a very high level of education (which tend to be college towns), or a very low level of diversity and a very low level of education (which is a huge swath of the rural Midwest). It’s not as effective as a deep dive into actual voter files that the pros have access to, but for an amateur like me, it seems to be a great way of sussing out how different races and different socio-economic classes performed, in terms of both turnout and how the percentage of their votes broke down.

(“Quintile” here means rank-ordering all the counties, and then grouping as many counties as it takes to equal one-fifth of the nation’s population. That’s better than simply using one-fifth of the nation’s 3,143 counties and county equivalents in each quintile, since there are a handful of multimillion-person counties, and hundreds of counties whose population is numbered only in the four digits. The “very high” diversity quintile is 0.9 to 41.1 percent white, followed by “high” at 41.2 to 57.6 percent white, “average” at 57.7 to 71.3 percent white, “low” at 71.4 to 84.1 percent white, and “very low” at 84.1 to 99.8 percent white. The “very high” education quintile is 78.8 to 37.7 percent college-educated, followed by “high” at 37.6 to 30.9 percent college-educated, “average” at 30.8 to 27.0 percent college-educated, “low” at 26.9 to 19.8 percent college-educated, and “very low” at 19.7 to 1.9 percent college-educated. These figures come from the Census Bureau’s 2011-15 American Community Survey, which is the most recent information available that encompasses all counties regardless of size.)

This is going to be a very chart-heavy article, and for data fans, the charts really do tell a fascinating story. But I’ll highlight some key takeaways for the benefit of the more verbally-inclined among us.

Comparison of 2016 and 2012 percentages

The more racially diverse and the more educated a county is, the more likely that county is to go Democratic. To start with, take a look at the chart above. In almost every box, the Dem percentages go from higher to lower as you go from left to right (i.e. more to less educated), and from higher to lower as you go from top to bottom (i.e. more to less diverse). The most solidly Democratic counties, with a 65 percent Hillary Clinton vote share or more, are the ones where there’s a synergy between both race and education, and are almost entirely in metro areas (though many of them are inner-ring suburbs, in addition to the cities themselves). The most solidly Republican counties, with a 35 percent Clinton vote share or less, have a dearth of both non-white voters and college-educated voters. The ones in the other corners tend to be the swingy regions (either the ones that are diverse but poorly-educated, which are usually rural agricultural areas in the south or west, or the ones that are mostly white but well-educated, which are usually college towns or outer suburbs).

(You might notice there’s a small aberration for the High Diversity/Average Education bucket. As best as I can tell, that’s because this bucket is dominated by Maricopa County, Arizona (where Phoenix is), which by itself has around 4 million people. Maricopa County, where a lot of the white residents are retirees and a lot of the Latino residents don’t vote, has tended to vote more right-wing than its demographic profile would imply, though it definitely moved more in line with its cohort in 2016, with Clinton losing only 48-45.

If you look at the disparity between the 2016 and 2012 numbers, you’ll also see some important trends. The upper-left quadrant counties are the parts of the country where Clinton actually ran ahead of Barack Obama’s percentages. (Think the Orange Counties and Cobb Counties of the country, as well as, say, the Seattles and Washington, DCs.) The swingy ones in the middle saw Clinton get a lower percentage than Obama, but also Donald Trump get a lower percentage than Mitt Romney, thanks to the much larger third-party share this year. And the lower-right quadrant counties saw a very large erosion of the Democratic vote share plus GOP gains, with the most extreme version in the very low diversity/very low education bucket, where Obama got 36 percent in 2012 but Clinton got only 25 percent in 2016. So, in other words, we saw the two ends of the electorate, which were already polarized, pulling even further apart over the last four years.

If you’re wondering how far back this trend goes, I calculated correlations between Dem vote share and white percentage in each county, and Dem vote share and college-educated percentage, going all the way back to 1984. The level of correlation with race was pretty consistent even in the 1980s up until 2008, but has dramatically pulled further apart since then. And the education trend is even more startling: in the 80s and 90s, more education was negatively correlated with voting Democratic (i.e. if you were college-educated, you were more likely to vote Republican), and even in the 00s, education had very little relationship with how you’d vote. But it suddenly accelerated in 2016 to nearly as robust a predictive relationship as you see with race.

2016: correlation w/non-Hispanic white: -0.59; w/ college education: 0.43

2012: non-Hispanic white: -0.45; college education: 0.29

2008: non-Hispanic white: -0.36; college education: 0.30

2004: non-Hispanic white: -0.33; college education: 0.20

2000: non-Hispanic white: -0.38; college education: 0.05

1996: non-Hispanic white: -0.40; college education: -0.07

1992: non-Hispanic white: -0.34; college education: -0.15

1988: non-Hispanic white: -0.24; college education: -0.11

1984: non-Hispanic white: -0.30; college education: -0.11

(If you’re not familiar with how correlation works, the values range from 1 to -1. 1 would mean a completely consistent relationship, i.e. if you ordered all counties from most Dem to most GOP, and most college-educated to least college-educated, the matches would be identical. -1 would be a totally inverse relationship. 0 would reflect no relationship at all; in other words, if you plotted the data, it would just look like a random spray. So, here, you can see that from 1984 to now, the relationship between how white your county is, and how likely you are to not vote Democratic started out somewhat noteworthy and got much stronger. How educated your county is switched from being very mildly predictive of not voting Democratic, to currently being pretty strongly predictive of voting Democratic.)

	‘16	‘12	‘08	‘04	‘00	‘96	‘92	‘88	‘84
VH/VH	D+28	D+22	D+20	D+20	D+18	D+14	D+16	D+13	D+13
VH/H	D+27	D+26	D+23	D+20	D+23	D+18	D+13	D+13	D+13
VH/A	D+19	D+15	D+13	D+12	D+11	D+8	D+7	D+5	D+4
VH/L	D+15	D+12	D+9	D+8	D+8	D+8	D+6	D+6	D+8
VH/VL	D+10	D+9	D+6	D+3	D+5	D+5	D+4	D+5	D+6
H/VH	D+19	D+12	D+12	D+10	D+8	D+5	D+5	D+3	D+4
H/H	D+14	D+10	D+10	D+9	D+8	D+5	D+5	D+2	D+3
H/A	D+2	R+2	R+3	R+3	R+4	R+4	R+4	R+3	R+2
H/L	D+5	D+7	D+6	D+5	D+5	D+5	D+3	D+3	D+4
H/VL	R+7	R+5	R+7	R+6	R+5	R+2	R+1	D+0	D+1
A/VH	D+10	D+6	D+6	D+5	D+3	D+1	D+1	R+0	R+1
A/H	D+2	D+1	D+1	D+0	R+0	R+3	R+4	R+5	R+4
A/A	D+1	D+1	D+1	D+0	R+0	R+2	R+2	R+0	D+1
A/L	R+8	R+6	R+6	R+5	R+5	R+3	R+3	R+2	R+2
A/VL	R+19	R+15	R+16	R+12	R+9	R+4	R+3	R+3	R+4
L/VH	D+6	D+3	D+3	D+4	D+3	D+1	D+1	D+1	D+1
L/H	R+2	R+4	R+2	R+3	R+2	R+2	R+3	R+2	R+2
L/A	R+8	R+6	R+6	R+5	R+5	R+5	R+5	R+4	R+4
L/L	R+13	R+9	R+8	R+7	R+6	R+5	R+5	R+4	R+4
L/VL	R+23	R+18	R+16	R+12	R+10	R+6	R+4	R+3	R+4
VL/VH	R+3	R+4	R+3	R+3	R+5	R+3	R+3	R+1	R+3
VL/H	R+7	R+6	R+5	R+5	R+7	R+6	R+7	R+6	R+7
VL/A	R+13	R+9	R+7	R+7	R+7	R+5	R+4	R+2	R+2
VL/L	R+16	R+10	R+9	R+7	R+6	R+4	R+4	R+2	R+3
VL/VL	R+25	R+15	R+12	R+9	R+8	R+4	R+2	R+2	R+3

Let’s look at one more chart, which is kind of complicated-looking, but helps make the correlation stuff I was talking about more concrete. This chart tracks the Partisan Voting Index of each bucket across the last nine elections. The counties in the upper-left quadrant buckets (high diversity and high education) were always the most Dem-friendly but also moved sharply in the Dem direction over the decades—for instance, from D+13 in 1984 to D+28 last year in the very high/very high bucket. The lower-left quadrant (high diversity but low education) counties moved only somewhat in the Dem direction, since the race and education cross-currents fought each other (but race is a bit stronger); the very high/very low bucket went from D+6 in 1984 to D+10 last year. (For purposes of this chart, the description left of the slash is always diversity, and the description right of the slash is always education.)

The upper-right quadrant has stayed almost entirely stationary over the years, especially in the very low diversity/very high education bucket, starting at R+3 in 1984 and staying R+3 in 2016. What may be most interesting, though, is if you look down the entire “very low diversity” column in the 80s or even 90s, there isn’t much difference between the different education levels; the mostly-white counties are all mildly GOP-leaning regardless of education, which back then, wasn’t much of a factor. For example, in 1984, very low diversity/very high education was R+3, very low diversity/high education was R+7, very low diversity/average education was R+2, very low diversity/low education was R+3, and very low diversity/very low education was R+3. In 2016, though, education became a huge factor: very low diversity/very high education was still R+3, very low diversity/high education was R+7, very low diversity/average education was R+13, very low diversity/low education was R+16, and very low diversity/very low education was a terrifying R+25.

(Partisan Voting Index, or PVI, developed by Charlie Cook, is the most useful tool for doing apples-to-apples comparison of presidential election results in particular CDs, states, or counties over time. The number doesn’t reflect the margin of victory or how much the percentage changed from one year to the next; it reflects the difference between the Dem (or GOP) vote share in that county vs. the Dem (or GOP) vote share nationwide. This is especially helpful when your comparison includes elections where there was a large third-party effect. Here’s an example: suppose a county gave Clinton 49 percent of the vote in 2016, and gave Obama 52 percent of the vote in 2012. You might say Clinton did worse in that case, but using PVI, she didn’t do relatively worse; in both elections, the PVI was D+1 (because Clinton did 1 pt. better than her 48 percent nationally, and Obama did 1 pt. better than his 51 percent nationally).

The total number of 2012 and 2016 votes in each bucket, and the percent change between the years

There weren’t significant turnout disparities between the different buckets. A lot of people, in the immediate aftermath of the election, seemed to assume that turnout was off in Democratic-leaning counties, while it was sky high in Republican counties. It seemed like an easy answer, but also one that seemed to satisfy some people’s ideological preferences (i.e. people bored with or angry at Clinton stayed home, and a different candidate would have driven more turnout). There were also a few blue counties where the total number of votes did genuinely decline—most notably Wayne County, Michigan, where there were 783,000 votes in 2016 and 818,000 votes in 2012, which is particularly noteworthy since that decline is similar to the amount that Clinton lost Michigan by. One important thing to keep in mind, though, is that Wayne County lost a corresponding amount of population during that period too, with people steadily emptying out of Detroit, mostly for the adjacent suburban counties.

However, if you look at the various buckets in the chart above and how much the gain in votes from 2012 to 2016 was in each one, that wasn’t really the case. The biggest percentage gains tended to be in buckets that were already blue, and getting bluer. For instance, the largest gain, 10 percent, was in the very high diversity/low education bucket, not the very bluest one but one where Clinton and Obama both got 63 percent. (Counties in this bucket include Miami-Dade County, Florida, Riverside County, California, and Bexar County, Texas, where San Antonio is.) The second largest gain was in the very high diversity/very high education bucket, which is mostly concentrated in the Bay Area and the Atlanta metro area.

By contrast, some of the smallest gains tended to be in the counties that are also the reddest (and ones reddening the fastest too): the very low diversity/very low education and very low diversity/low education buckets, which gained only 4 percent over the four years. Only two buckets did worse than that: the high diversity/low education bucket (where, not coincidentally, Wayne County, Michigan, is one of the most populous examples) and especially the high diversity/very low education bucket, the only one that actually had fewer votes overall in 2016, though only by around 4,000.

(The best explanation I can come up with for the failure in the high diversity/very low education bucket is that those counties are disproportionately found in rural eastern North Carolina, which has a large black population, though not to the extent of the Mississippi Delta or Alabama’s Black Belt, which fall in the “very high diversity” bucket. These North Carolina counties were also extremely hard-hit by flooding from Hurricane Matthew shortly before the election, so a lot of voters there were entirely displaced or otherwise preoccupied.)

But for most of the buckets all gained votes at a pretty comparable rate, usually around a 6 or 7 percent gain (which is entirely consistent with the 6 percent national gain from 2012 to 2016: 128.8 million in 2012 to 136.7 million in 2016).

The ratio between the total number of votes, and total citizen voting-age population, in each bucket

Another way of looking at this question, as seen in the chart above, is by looking directly at turnout as a share of population in 2016, rather than by comparing 2012 votes to 2016 votes. Again, this doesn’t show turnout in the dark red counties swamping turnout in the dark blue counties. If anything, you’ll notice slightly higher turnout percentages in the upper-left quadrant than in the lower-right quadrant (i.e. 64 percent in the very high diversity/very high education bucket vs. 57 percent in the very low diversity/very low education bucket).

You’ll notice that turnout was highest in the upper right quadrant, in the low diversity/high education counties. It was lowest in high diversity/low education counties. That makes a great deal of sense, since turnout has historically been correlated with affluence. Having the time to inform yourself about the issues, and in fact the time needed to vote, if you’re in a locale where there tend to be lines is, essentially, something of a luxury good. As you can see in the chart, turnout progresses from bucket to bucket nearly as consistently as on the Democratic vote share chart we looked at earlier … except it moves in partially the opposite direction, as turnout goes up the whiter the county is, in addition to how well-educated it is.

(Note that when I say “population” here, I’m referring to citizen, voting-age population, rather than everyone, so that turnout doesn’t seem disproportionately low in counties where there are a lot of non-citizens or a lot of people under 18 which, not coincidentally, tend to be in the same counties). Turnout is often expressed as a percentage of votes out of registered voters, rather than CVAP, and ideally I’d use that data point instead, but registration data is difficult to collect and some places (most notably North Dakota, where there simply is no registration) don’t even make that data available.)

	‘16	‘12	‘08	‘04	‘00	‘96	‘92	‘88	‘84
VH/VH	2.2	2.1	2.1	2.0	2.0	2.0	2.1	2.1	2.1
VH/H	1.9	2.0	1.9	2.0	2.0	2.1	2.0	2.2	2.3
VH/A	5.7	5.6	5.6	5.6	5.8	5.7	6.0	6.3	6.4
VH/L	3.0	2.9	2.9	2.8	2.8	2.8	2.8	2.9	2.9
VH/VL	2.9	2.8	2.8	2.7	2.8	2.8	2.8	2.8	2.8
H/VH	5.6	5.5	5.4	5.3	5.4	5.3	5.3	5.4	5.2
H/H	5.5	5.5	5.4	5.5	5.5	5.5	5.7	5.8	5.8
H/A	3.8	3.8	3.6	3.6	3.3	3.2	3.3	3.1	2.9
H/L	2.8	2.9	2.8	2.8	2.7	2.7	2.7	2.7	2.8
H/VL	1.7	1.8	1.8	1.8	1.8	1.9	1.9	1.9	2.0
A/VH	6.8	6.7	6.5	6.6	6.6	6.5	6.5	6.4	6.3
A/H	5.3	5.3	5.2	5.3	5.3	5.2	5.3	5.2	5.1
A/A	2.9	2.9	2.9	3.0	3.0	3.1	3.2	3.2	3.2
A/L	3.5	3.5	3.4	3.4	3.3	3.3	3.3	3.2	3.2
A/VL	2.4	2.4	2.3	2.4	2.4	2.5	2.4	2.4	2.5
L/VH	5.3	5.3	5.1	5.3	5.3	5.2	5.1	5.2	5.0
L/H	4.9	4.9	4.8	4.9	4.9	4.9	4.8	4.8	4.6
L/A	4.1	4.0	3.9	4.0	3.9	3.8	3.7	3.5	3.4
L/L	4.3	4.3	4.2	4.3	4.2	4.1	4.1	3.9	3.9
L/VL	3.3	3.4	3.3	3.4	3.4	3.5	3.5	3.4	3.5
VL/VH	2.3	2.3	2.2	2.2	2.1	2.0	1.9	1.9	1.7
VL/H	2.9	2.9	2.9	3.0	2.9	2.8	2.7	2.6	2.5
VL/A	2.4	2.4	2.4	2.5	2.4	2.4	2.4	2.4	2.3
VL/L	6.4	6.5	6.4	6.8	7.0	7.1	7.0	7.0	7.2
VL/VL	8.0	8.2	8.2	8.9	9.2	9.5	9.5	9.6	10.0

Let’s look at one last chart here: this shows what percentage of the total number of votes, nationwide, are found within each bucket over the years. As an example, look at the very high diversity/very high education bucket over the years. In 2016, there were 136.7 million votes total, and there were almost exactly 3 million votes in the very high diversity/very high education bucket; that’s 2.2 percent of the national total. In 1984, there were only 92.4 million votes, and 1.9 million votes in the very high diversity/very high education bucket, which is 2.1 percent of the national total. It’s one category that’s been remarkably consistent over the decades.

In fact, most of the categories have been very stable, their overall share changing only a fraction of a percent over 30 years. The biggest drops are the very low diversity/very low education and very low diversity/low education buckets, in the dark-red, lower-right corner. Very low diversity/very low education fell from 10.0 percent of the nation in 1984 to only 8.0 percent in 2016, and very low diversity/low education fell from 7.2 percent in 1984 to 6.4 percent in 1984. That’s unsurprising, considering that these rural areas, mostly in the Midwest, are generally stagnant or even slowly losing population each year, while the nation’s metro areas continue to gain rapidly. The biggest gainer was the high diversity/average education bucket, going from 2.9 percent of the nation in 1984 to 3.8 in 2016; again, that’s largely because of how thoroughly the fast-growing Maricopa County, Arizona, dominates that particular bucket.

Even those are not big changes in the grand scheme of things, but it does show that the low diversity/low education parts of the country are becoming a smaller and smaller share of the overall vote even as they become much redder. So, again, the Dem-friendly areas were not swamped by the rural red areas in 2016, and in fact the whole matrix is changing at an almost geologically slow pace.

Instead of turnout being the problem, the real killer was Obama → Trump switchers in rural counties. So, we haven’t seen any major turnout shifts. Turnout, in terms of the number of 2012 voters vs. 2016 voters, didn’t fall off in blue areas and didn’t shoot up in red areas. (In fact, blue areas, taken as a whole, did slightly better, though that’s largely an artifact of metro areas, which tend to be bluer and growing at a faster clip in terms of overall population than rural areas, which also tend to be redder.)

Instead, what we’ve seen is a sharp falloff in the Democratic vote share in those rural red areas, especially in the 2012 to 2016 period. (Though it’s been a steady decline over the decades. The 38 percent that Walter Mondale got in the very low diversity/very low education counties is still higher than the 36 percent that Obama got in 2012, to say nothing of the 25 percent that Clinton got in 2016—even while Mondale was being nuked nationwide.) In other words, it wasn’t a turnout failure as much as it was a persuasion failure.

Since there wasn’t a large change in the total number of votes in the VL/VL bucket, it’s a safe conclusion there was no influx of newly-activated voters. In fact, if you randomly grab any small, rural county in the Midwest and look at its 2012 and 2016 numbers, you’re likely to see the same pattern at a microscopic level. The overall number of votes was generally around the same, maybe an increase of a few hundred more out of, say, 4,000 total votes … but a county that Obama lost “only” 40-60 instead becomes one that Clinton lost 30-40. Taken individually, it’s purely “meh,” but when you multiply that by hundreds of counties, many of which are in key swing states, you’ve swung enough votes to swing Michigan, Ohio, Pennsylvania, and Wisconsin.

One of the most interesting and memorable postmortems after the 2012 election was by Real Clear Politics’ Sean Trende, who posited that there was a large bloc of “missing white voters” who were basically working-class, poorly-educated voters feeling alienated from the political system as a whole. By overlaying turnout numbers with demographic data, he found they seemed to be especially located in an arc of northern states, especially in areas where there tend to be a lot of secular/unchurched voters and, in ‘92/’96, a lot of Ross Perot voters.

One of my first reactions on seeing the post-2016 map was “holy shit, Trump actually found the missing white voters!” But having looked in more depth at the numbers over the months, it doesn’t seem like that’s the case. There was no surge of new, off-the-grid voters added to the rolls. (It’s mathematically possible that there was a surge of “missing white voters” and an equal counter-surge of dissatisfied Dems staying home, but it all seems a little too convenient that in rural county after county where the overall vote totals hardly changed from ‘12 to ‘16, that the number of newly-activated Trumpeters was always roughly equal to the number of Dems opting out.)

Instead, it was the same white voters who’ve been there in plain sight all along; instead, around 1 out of 10 of them in rural counties found Trump uniquely appealing and decided to switch. The real question for coming years is whether that’s permanent, or just a temporary, candidate-specific aberration. (The Perot connection, oddly enough, is one thing that gives me hope that Trumpomania was a one-time thing; focus groups with Obama/Trump voters suggest that what many of them found most appealing was the heterodox ideology and overly-simplistic, problem-solving outsider approach that he was putting forward, i.e. that politics should be “more like business,” very similar to the Perot campaign of 1992. The last time that happened, it didn’t turn into a sustainable movement; it was more a cult of personality that disappeared into the mists, along with Perot himself.)

The solution, going forward, isn’t either/or, but both. One of the main current debates in the blogosphere and the Democratic Party writ large is whether the path out of our shared predicament is by trying to consolidate our new gains in the formerly-Republican, diverse, and well-educated suburbs of the Sun Belt (the high diversity/high education buckets), or by trying to win back the blue-collar Midwesterners (in the low diversity/low education buckets) who were lured by the Trump siren song but may be responsive to something more populist.

It’s kind of a mind-numbing debate though, since we don’t need to do one or the other; in fact, we very much need to do both, and there’s nothing incompatible about them. Given how narrow the miss in 2016 was (around 100,000 votes spread out across three key states), we don’t even need to do much of either one, but we need to push ahead with both: the former will help us get over the top in Florida and North Carolina (and maybe even Georgia and Arizona), and the latter will help us get back on our feet in Michigan, Ohio, Pennsylvania, and Wisconsin.

I’d somewhat lean toward focusing on the high diversity side of the chart. That's largely because those are already moving in our direction in terms of Democratic vote share, and those are growing parts of the country as seen in the superior turnout numbers there compared with the low diversity/low education quadrant. In general, whether you’re marketing widgets or a presidential candidate, you want to target a growing segment of the population, as well as one that’s more culturally disposed to what you’re selling. (On top of that, there are efficiency advantages associated with get-out-the-vote in denser areas; it’s simply a lot more effective to canvass in areas where people are close together, instead of places where you have to travel significantly just to get from house to house.)

However, there are several reasons that we can’t just write off the residents of the rural parts of the Midwestern states. One is Electoral College math; the swing states in the Midwest are much more heavily reliant on the low diversity/low education population than the ones in the south or west. Here are a few examples: in Florida, the very low diversity/very low education bucket encompasses only 1.0 percent of the population and the very low diversity/low education bucket is another 1.5 percent. In North Carolina, the very low diversity/very low education bucket is 5.2 percent of the state, while the very low diversity/low education bucket is another 3.4 percent.

By contrast, in Wisconsin, the very low diversity/very low education bucket is 16.7 percent of the state, and the very low diversity/low education bucket is another 19.7 percent. In Pennsylvania, very low diversity/very low education is 15.2 percent of the state, and very low diversity/low education is 18.4 percent. Ohio might be the most extreme example: 24.8 percent of the state is in the very low diversity/very low education bucket, and 14.4 percent more are in the very low diversity/low education bucket. Write off the rural parts of those states, and you’ve essentially junked the decisive share of the vote.

The other problem is the races that are districted and aren’t decided statewide: House races, and the fight over state legislatures. Thanks to the nefarious nexus of gerrymandering and self-clustering, Democratic votes tend to be concentrated in fewer districts in the cities and inner suburbs. The exurbs and rural areas, which tend to be less diverse and less educated, also get disproportionate power in the House and state legislatures because the Dem-friendly demographics are so geographically concentrated, leaving the rest of the seats spread out across what’s left of the state. If we’re going to retake the House, it means playing in CDs in the lower-right quadrant.

Of course, it’s not hard to do that at the House level, where different candidates can take different approaches that fit their districts; in fact, we’re already doing that pretty well in the current special elections, with Jon Ossoff running a technocratic campaign in the high diversity/high education suburbs of GA-06, and Rob Quist running a populist campaign in the low diversity/low education MT-AL. (It’s a little harder at the presidential level, where you have to find one charismatic-yet-blank-slate person who’s all things to all people in all demographics. Even then, though, you can still run differently-pitched ads in, say, the Philadelphia media market vs. the Johnstown media market.)

In short, there’s no one-size-fits-all solution (and anyone who told you there was was lying to you). It’s important to understand what kind of candidate, and what kind of rhetorical approach, works in what part of the country. And with race and education becoming even more determinative than before in terms of what works where, it’s important to understand the demographics of wherever you’re trying to appeal to—without, of course, simply saying “demographics is destiny” and washing your hands, without doing the actual grunt work of showing up and GOTVing.

APPENDIX

You didn’t think that was it, did you? You might still be wondering “well, what bucket does the county where I live fall into?” Let’s talk about that too: we’ve got maps, as well as hard data about each of the 25 buckets. We’ll give an overview of each category, with diversity as the first level of sort and education as the second. (I.e. we’ll talk about the “very high diversity” ones first, from “very high education” to “very low education,” then switch to “high diversity,” and so on.)

I didn’t want to crash your browser with 25 different maps, so I’ve combined the maps somewhat for a total of 10: within each diversity column, there’s one map for the very high/high/average education rows, and then another for the low/very low education rows. Within each map, you’ll see four colors. The dark blue counties are the ones that were won by both Obama in ‘12 and then Clinton; the red ones went for Romney and Trump. The light blue ones flipped from Romney to Clinton; appropriately, the orange ones went from Obama to Trump.

Very high diversity; very high/high/average education

VERY HIGH DIVERSITY/VERY HIGH EDUCATION

Total population: 7,207,912

7 counties; avg. population 1,029,702

Largest examples: Santa Clara, CA (San Jose); Alameda, CA (Oakland); Fulton, GA (Atlanta)

2016: 74.2% C/19.9% T/5.9% O (total: 3,007,070; 2.2% of nationwide)

2012: 72.5% O/25.3% R/2.1% O (total: 2,741,384; 2.1% of nationwide)

This bucket contains only a few counties, each of which has a large population. These are all very affluent areas, but yet all but one of them are very dark blue, including Washington, DC, the most Dem-friendly jurisdiction in the nation. The one exception is Fort Bend County, Texas, in Houston’s suburbs, which Romney won in ‘12 but Clinton narrowly won in ‘16.

VERY HIGH DIVERSITY/HIGH EDUCATION

Total population: 7,419,120

10 counties; avg. population 741,912

Largest examples: Kings, NY (Brooklyn); Honolulu, HI; Prince George’s, MD (DC suburbs)

2016: 74.7% C/20.5% T/4.8% O (total: 2,654,964; 1.9% of nationwide)

2012: 76.9% O/21.5% R/1.6% O (total: 2,543,247;  2.0% of nationwide)

This bucket is pretty similar to the very high diversity/very high education bucket above it, with a mix of big cities and inner suburbs, all of which are dark blue. In fact, it has a slightly higher Dem percentage than the very high diversity/very high education bucket, thanks to a) the absence of swingy Fort Bend Co., TX, and b) the presence of Prince George’s County, MD, a majority-African-American suburban county to the southeast of Washington, DC, with nearly 900,000 residents that gave Clinton 88 percent of the vote.

VERY HIGH DIVERSITY/AVERAGE EDUCATION

Total population: 23,162,344

12 counties; avg. population 1,930,195

Largest examples: Los Angeles, CA; Harris, TX (Houston); Dallas, TX

2016: 67.2% C/32.8% T/4.8% O (total: 7,828,511; 5.7% of nationwide)

2012: 65.8% O/32.5% R/1.7% O (total: 7,275,591; 5.6% of nationwide)

This is one of the most populous buckets in the nation, in large part thanks to the presence of Los Angeles County, CA, which has about 10 million residents, by far the largest county in the nation. On top of that, the nation’s third most-populous county, Harris, TX (location of Houston) is in here too. Further down the list, this bucket also contains Queens in New York City, Ft. Lauderdale, and Memphis, so it’s also heavy on the big cities.

Very high diversity; low/very low education

VERY HIGH DIVERSITY/LOW EDUCATION

Total population: 11,564,169

24 counties; avg. population 481,840

Largest examples: Miami-Dade, FL; Riverside, CA; Bexar, TX (San Antonio)

2016: 62.7% C/32.8% T/4.5% O (total: 4,120,296; 3.0% of nationwide)

2012: 63.3% O/35.4% R/1.3% O (total: 3,760,802; 2.9% of nationwide)

One thing you’ll notice is that the number of counties per bucket is starting to go up, as we get into more rural counties with smaller populations. Another thing is that we’re finally getting down into a few counties that went for Trump (including Nueces Co., TX, where Corpus Christi is), though the large majority are still blue counties. For the most part, these are large cities with big immigrant populations, though it also includes a few rural counties with Latino or black majorities.

VERY HIGH DIVERSITY/VERY LOW EDUCATION

Total population: 12,933,239

173 counties; avg. population 74,758

Largest examples: San Bernardino, CA; Bronx, NY; Fresno, CA

2016: 58.0% C/37.0% T/5.0% O (total: 3,904,971; 2.9% of nationwide)

2012: 60.3% O/38.1% R/1.6% O (total: 3,635,933; 2.8% of nationwide)

Here, we’ve switched mostly to rural counties (though there are a few big urban areas, like the Bronx). More specifically, it’s the rural counties with Latino, black, or Native American majorities, most of which are pretty impoverished. If you look at the map, you can see the two arcs pretty clearly: the Hispanic counties along the border in Texas, extending into New Mexico and then skipping over to California’s Central Valley, and the black counties following the Delta and black belt across Mississippi and Alabama into South Carolina. These are traditionally low-turnout areas, though Obama was able to significantly change that and Clinton was able to continue the momentum, at least in the Latino-majority counties.

High diversity; Very high/high/average education

HIGH DIVERSITY/VERY HIGH EDUCATION

Total population: 17,573,925

26 counties; avg. population 675,920

Largest examples: Orange, CA; New York, NY (Manhattan); Fairfax, VA (DC suburbs)

2016: 65.8% C/28.2% T/6.0% O (total 7,698,336; 5.6% of nationwide)

2012: 63.0% O/35.0% R/2.0% O (total 7,107,965; 5.5% of nationwide)

This bucket contains much of the nation’s money and power; it has Manhattan, and a little further down the list, San Francisco. It also has Fairfax Co., Virginia, and Montgomery Co., Maryland, where the people in power in Washington, DC, tend to live rather than the city itself, and other knowledge economy hubs like Travis Co., Texas (Austin), and Suffolk Co., Massachusetts (Boston). It has two of the most prominent Romney-to-Clinton counties (Orange Co., California, and Cobb Co., Georgia, in Atlanta’s suburbs) … and it even has one dark-red county (Brazos Co., Texas, where Texas A&M is).

HIGH DIVERSITY/HIGH EDUCATION

Total population: 17,908,389

20 counties; avg. population 895,419

Largest examples: Cook, IL (Chicago); San Diego, CA; Palm Beach, FL

2016: 62.0% C/32.6% T/5.3% O (total 7,576,361; 5.5% of nationwide)

2012: 61.1% O/37.3% R/1.6% O (total: 7,028,137; 5.5% of nationwide)

This bucket is dominated by Cook County, Illinois, where Chicago is (which is the second-most populous county in the nation, with around 5 million people), but it also contains a number of middle-class but not-quite-so-affluent suburbs compared to the high diversity/very high education bucket above, like Ventura County, California, and Gwinnett County, Georgia (which, like neighboring Cobb Co., went Romney to Clinton). It also includes a number of smaller cities around the south that are cultural anchors for their broader surroundings, like Davidson Co., Tennessee (Nashville), Jefferson Co., Alabama (Birmingham), and even Chatham Co., Georgia (Savannah).

HIGH DIVERSITY/AVERAGE EDUCATION

Total population: 12,890,740

15 counties; avg. population 859,353

Largest examples: Maricopa, AZ (Phoenix); Tarrant, TX (Ft. Worth); Sacramento, CA

2016: 49.4% C/44.4% T/6.2% O (total 5,253,526; 3.8% of nationwide)

2012: 49.5% O/48.7% R/1.7% O (total 4,856,156; 3.8% of nationwide)

This bucket is heavy on the cities of the Sun Belt that aren’t quite as knowledge economy-focused; it starts with Phoenix, and if you go a little further down the list, it also includes places like Tampa, Jacksonville, and Tucson (and looking further north, it also includes Milwaukee, which has a larger African-American population than you might expect). As I speculated earlier, this bucket is a little disproportionately red compared with its neighbors, owing largely to how big an influence Maricopa County is within this particular group.

High diversity; Low/very low education

HIGH DIVERSITY/LOW EDUCATION

Total population: 9,619,709

50 counties; avg. population 192,394

Largest examples; Clark, NV (Las Vegas); Wayne, MI (Detroit); Passaic, NJ (Paterson)

2016: 53.0% C/42.0% T/5.0% O (total 3,810,864; 2.8% of nationwide)

2012: 57.7% O/40.9% R/1.4% O (total 3,719,821; 2.9% of nationwide)

The High diversity/Low education bucket has a couple of major cities that are mostly on the wrong side of the information economy divide: manufacturing-oriented Detroit and tourist/service-sector Las Vegas. It also includes some smaller blue-collar cities with large minority populations like Norfolk, Virginia, Shreveport, Louisiana, and Pueblo, Colorado (which has a large Latino population), and some rural swaths of Virginia and Georgia. Here we’re finally starting to see some counties where Clinton lost a lot of ground from Obama (including, again, Pueblo Co., Colorado, which you’ll notice is orange on the map, along with some of its rural neighbors).

HIGH DIVERSITY/VERY LOW EDUCATION

Total population: 6,387,045

185 counties; avg. population 34,524

Largest examples: Stanislaus, CA (Modesto); Jefferson, TX (Beaumont); Yakima, WA

2016: 41.9% C/54.1% T/4.0% O (total: 2,346,858; 1.7% of nationwide)

2012: 46.3% O/52.4% R/1.3% O (total: 2,350,151; 1.8% of nationwide)

Here we’re seeing mostly rural counties, and if you look at the map, they’re mostly red counties. The problem here is that these are counties that tend to be around half white, half non-white, but turnout tends to be low among the non-white half, and the blue-collar white plurality tends to be reflexively conservative. These counties tend to be in close geographic proximity to the Very High Diversity/Very Low Education counties, but not as heavily-minority so also not as blue (i.e. further away from the Rio Grande, or from the Mississippi Delta).

Average diversity; Very high/high/average education

AVERAGE DIVERSITY/VERY HIGH EDUCATION

Total population: 19,108,637

37 counties; avg. population 516,449

Largest examples: King, WA (Seattle); Nassau, NY (NYC suburbs); Hennepin, MN (Minneapolis)

2016: 57.0% C/36.1% T/6.8% O (total 9,283,103; 6.8% of nationwide)

2012: 56.4% O/41.6% R/2.0% O (total 8,642,827; 6.7% of nationwide)

This is one of the nation’s most populous buckets, containing many of the nation’s affluent suburban counties, like Nassau Co. on Long Island, Fairfield Co., Connecticut, Lake and DuPage Counties, Illinois, and St. Louis Co., Missouri (which, confusingly, doesn’t contain the city of St. Louis proper). It also contains a few very large counties that have both a major knowledge sector city in them plus a lot of their suburbs, including King Co., Washington, Hennepin Co., Minnesota, Multnomah Co., Oregon (location of Portland), and Wake Co., North Carolina (where Raleigh is). Despite not having above-average levels of minority residents, these counties still swung noticeably in Clinton’s direction thanks to a well-educated, upper middle-class outlook. There are a few outliers: Nassau Co. swung away from Clinton, though she still won, and Clinton still lost the wealthy outposts outside of Dallas, Collin, and Denton Cos., despite them moving sharply in her direction.

AVERAGE DIVERSITY/HIGH EDUCATION

Total population: 15,895,849

37 counties; avg. population 429,617

Largest examples: Suffolk, NY (NYC suburbs); Franklin, OH (Columbus); Hartford, CT

2016: 50.2% C/44.1% T/5.6% O (total 7,310,481; 5.3% of nationwide)

2012: 52.5% O/45.9% R/1.6% O (total 6,862,103; 5.3% of nationwide)

This bucket starts out with a big outlier: Suffolk Co., which covers the eastern half of Long Island, is by far the most populous county in the nation to flip from Obama to Trump. At first glance, it doesn’t seem to have much in common with the rural Midwest, but if you look past the Hamptons toward its more mundane subdivisions, you’ll find a lot of New York Post-reading, Howard Stern-listening, blue-collar white guys who are prime Trumpenproletariat material. Much of the rest of this category, however, is mid-sized cities of regional importance, like Columbus, Cincinnati, and Louisville, along with some other northeastern suburbs with a bit of working-class flavor, like Delaware Co. outside of Philadelphia.

AVERAGE DIVERSITY/AVERAGE EDUCATION

Total population: 9,135,147

32 counties; avg. population 285,473

Largest examples: Cuyahoga, OH (Cleveland); Marion, IN (Indianapolis); Oklahoma, OK (Oklahoma City)

2016: 48.9% C/45.6% T/5.6% O (total 3,917,448;  2.9% of nationwide)

2012: 52.6% C/46.1% R/1.3% O (total 3,772,834; 2.9% of nationwide)

Here is literally the most middlebrow portion of the nation, and appropriately, the 2012 and 2016 vote shares are remarkably close to the nationwide totals. In addition to stolid Midwestern cities like Cleveland and Indianapolis, it also contains places like Wichita, Allentown, and Lafayette, Louisiana. Most amusingly, it contains the place whose very middlebrow-ness has been its defining characteristic for centuries: Peoria, Illinois, as in “will it play in Peoria?” (And, yes, Peoria has been used as a favored test market not just for theater but for almost everything else that can be marketed, thanks to its demographic averageness.)

Average diversity; Low/very low education

AVERAGE DIVERSITY/LOW EDUCATION

Total population: 11,155,052

83 counties; avg. population 134,398

Largest examples: Pierce, WA (Tacoma); Lee, FL (Ft. Myers); Providence, RI

2016: 40.5% C/53.9% T/5.6% O (total 4,740,529; 3.5% of nationwide)

2012: 45.3% C/53.1% T/1.5% O (total 4,488,433; 3.5% of nationwide)

This bucket has a lot of rural southern counties in it, though at the top of the list it has a few blue counties (not just Tacoma and Providence, but also Hampden Co., Massachusetts (Springfield), St. Clair Co., Illinois (E. St. Louis), and Lucas Co., Ohio (Toledo) that keep it from being too red overall. The common thread, though, tends to be declining industrial centers; in addition to the cities just mentioned, it also includes some smaller manufacturing towns in the south (like Biloxi, Mississippi, and Lake Charles, Louisiana), and one of the most alarming Obama-to-Trump flippers, Saginaw County, Michigan. There are also a few outlying Florida counties here like Lee Co. and St. Lucie Co., where Ft. Pierce is, which tend to attract a lot of working-class retirees from the Midwest.

AVERAGE DIVERSITY/VERY LOW EDUCATION

Total population: 8,223,882

266 counties; Avg. population 30,916

Largest examples: Polk, FL (Lakeland); Pinal, AZ (Phoenix suburbs); Onslow, NC (Jacksonville)

2016: 31.3% C/65.2% T/3.6% O (total 3,222,681; 2.4% of nationwide)

2012: 36.5% O/62.2% R/1.3% O (total 3,103,091; 2.4% of nationwide)

Now we’re really starting to get into rural America, with 266 counties in this bucket, most of which have a population under 25,000. These counties are disproportionately in the South, where there’s generally enough of an African-American population in these counties to move them into the “average” diversity column; the “low” and “very low” diversity rural counties instead tend to be in the Midwest. The one county here with a population over 500,000 is Polk County, Florida, which has both a lot of downscale retirees and much of the state’s agricultural activity, meaning a lot of immigrant farmworkers.

Low diversity; Very high/high/average education

LOW DIVERSITY/VERY HIGH EDUCATION

Total population: 13,593,045

48 counties; Avg. population 283,188

Largest examples: Middlesex, MA (Cambridge); Allegheny, PA (Pittsburgh); Oakland, MI (Detroit suburbs)

2016: 53.5% C/39.7% T/6.9% O (total 7,239,504; 5.3% of nationwide)

2012: 53.6% O/44.6% R/1.8% O (total 6,820,782; 5.3% of nationwide)

This bucket is disproportionately filled by the nation’s college towns, places like Madison, Boulder, Iowa City, Lawrence, Ithaca, Corvallis, and Pullman. College towns, naturally, have some of the highest education rates in the country (even though the Census figure I’m relying on only counts persons over 25 with degrees, rather than current college students; the faculty, associated industries, and assorted hangers-on all account for those numbers). It also includes some affluent mostly-white suburbs (like Marin Co., California, in the Bay Area, Montgomery Co., Pennsylvania near Philadelphia, and Johnson Co., Kansas outside Kansas City), and one county that fits in both categories (Middlesex Co., Massachusetts, which is mostly suburbs but also has Cambridge).

LOW DIVERSITY/HIGH EDUCATION

Total population: 14,264,780

58 counties; Avg. population 245,944

Largest examples: Salt Lake, UT; Erie, NY (Buffalo); Worcester, MA

2016: 44.8% C/46.1% T/9.2% O (total 6,759,158; 4.9% of nationwide)

2012: 47.3% O/50.8% R/1.9% O (total 6,320,335; 4.9% of nationwide)

One little detail that might jump out at you in this bucket is how large the third-party vote share is: 9.2 percent! However, the Low diversity/High education bucket isn’t a hotbed of Jill Stein voters. One quirk here is that Utah’s two most populous counties (Salt Lake and Utah Co., where Provo is) are both in this bucket; they’re both mostly white but are well-educated (with Salt Lake City one of the west’s cultural oases and Provo the home of Brigham Young Univ.). So, naturally, that means a disproportionate share of Evan McMullin votes in this bucket. Much of the rest of this bucket is midsize, middlebrow cities across the northeast and Midwest, including Buffalo, Rochester, Grand Rapids, and Des Moines, and further south, cities with universities like Knoxville and Asheville.

LOW DIVERSITY/AVERAGE EDUCATION

Total population: 11,632,978

58 counties; avg. population 200,568

Largest examples: Pinellas, FL (St. Petersburg); Snohomish, WA (Everett); Brevard, FL (Melbourne)

2016: 40.3% C/52.6% T/7.2% O (total 5,566,696; 4.1% of nationwide)

2012: 44.8% O/53.2% R/2.1% O (total 5,211,866; 4.0% of nationwide)

This bucket is centered around what you might call prosperous working-class suburbs, like Snohomish Co. outside of Seattle, Clark Co., Washington, outside of Portland, Summit Co., Ohio, outside of Cleveland (containing Akron, but largely suburban), and Anoka Co., Minnesota. The largest example in this bucket, though, is Pinellas County, Florida, which is one of the most populous counties in the country to go Obama to Trump, which was instrumental in the narrow Clinton loss statewide in Florida. Pinellas is, indeed, partly a bedroom community for the Tampa Bay area, but also has a lot of Midwestern retirees living near the beach.

Low diversity; Low/very low education

LOW DIVERSITY/LOW EDUCATION

Total population: 12,623,482

136 counties; avg. population 92,819

Largest examples: Macomb, MI (Detroit suburbs); Montgomery, OH (Dayton); Lancaster, PA

2016: 36.0% C/58.2% T/5.9% O (total 5,887,845; 4.3% of nationwide)

2012: 42.3% C/56.1% R/1.6% O (total 5,524,360; 4.2% of nationwide)

The Low diversity/Low education bucket has a lot of Midwestern and northeastern industrial towns and downscale suburbs in it: places like Lorain Co., Ohio (outside of Cleveland), Mahoning Co., Ohio (where Youngstown is), Berks Co., Pennsylvania (home of Reading), and Oneida Co., New York (Utica). Most distinctively, it has the very place that’s most closely associated in the conventional wisdom-spouting pundit’s mind with “Reagan Democrats:” Macomb County, Michigan, home of the prototypical non-college white Catholic unionized autoworker. (And also one of the largest Obama to Trump-flipping counties in 2016.) It also contains some counties in Florida that are places that people who used to live in places like Macomb County like to retire to, like Volusia Co. (Daytona Beach) and Pasco Co. (north of Tampa).

LOW DIVERSITY/VERY LOW EDUCATION

Total population: 10,921,175

257 counties; avg. population 42,494

Largest examples: Genesee, MI (Flint); Marion, FL (Ocala); Gaston, NC (Gastonia)

2016: 27.0% C/68.4% T/4.7% O (total 4,565,328; 3.3% of nationwide)

2012: 33.9% O/64.5% R/1.6% O (total 4,354,224; 3.4% of nationwide)

This bucket is mostly rural, with the most urban exception being Genesee County, Michigan, where Flint is located. Genesee is also one of the very few blue counties in this bucket; you might think that Flint’s residents were especially revved-up to vote in this election, but it’s worth noting that most of the people in this county don’t live in Flint proper, and they tend to be white working-class, similar to the residents of Macomb County just to their south. Here, we’re starting to move away from the South, except for some areas in the less diverse Appalachian-flavored uplands of Alabama and North Carolina. If you look closely, you’ll notice that this bucket contains some of the less enjoyable blue-collar towns in the west, like Redding, California, Kingman, Arizona, and Aberdeen, Washington (whose county, Grays Harbor County, went GOP for the first time in 2016 since 1928!).

Very low diversity; Very high/high/average education

VERY LOW DIVERSITY/VERY HIGH EDUCATION

Total population: 5,602,217

54 counties; avg. population 103,744

Largest examples: Waukesha, WI (Milwaukee suburbs); Rockingham, NH (Boston suburbs); Hamilton, IN (Indianapolis suburbs)

2016: 44.9% C/47.6% T/7.4% O (total 3,085,117; 2.3% of nationwide)

2012: 46.7% O/51.4% R/1.9% O (total 2,898,197; 2.3% of nationwide)

This is the bucket with the lowest population (though the poor turnout in the Very High Diversity/Very Low Education bucket saves it from having the lowest number of votes). It’s heavy on the affluent, conservative suburbs, including the most notorious one of all, crucial Waukesha County in Wisconsin. It also has Waukesha’s mega-churchy counterpart in Indiana, Hamilton County, and Rockingham County in New Hampshire, where people who want to work in Boston but flee Massachusetts’s taxes live. It also has some extra-white college towns in it, like State College, Ames, Missoula, and Burlington, Vermont.

VERY LOW DIVERSITY/HIGH EDUCATION

Total population: 7,672,214

82 counties; avg. population 93,563

Largest examples: Bucks, PA (Philly suburbs); Ada, ID (Boise); Hillsborough, NH (Manchester)

2016: 40.0% C/51.0% T/8.9% O (total 4,020,112; 2.9% of nationwide)

2012: 44.8% O/53.2% T/2.1% O (total 3,779,809; 2.9% of nationwide)

As with the Very Low diversity/Very High education bucket, this collection is heavy on the affluent suburbs and exurbs, most notably Bucks County to the north of Philadelphia. Other examples are St. Charles Co., Missouri, at the outer reaches of the St. Louis metro area, and Cumberland Co., Pennsylvania, outside of Harrisburg. Finally, it has some standalone small cities in the west that have a fair amount of commercial activity going on, like Fargo, Boise, and Bend, Oregon (though Boise and Bend have enough ex-Californians in them that they’re really more like white flight exurbs for the Los Angeles area).

VERY LOW DIVERSITY/AVERAGE EDUCATION

Total population: 6,691,187

96 counties; avg. population 69,699

Largest examples: Spokane, WA; Greene, MO (Springfield); Washington, PA (Pittsburgh suburbs)

2016: 35.4% C/57.1% T/7.5% O (total 3,261,534; 2.4% of nationwide)

2012: 42.0% O/55.8% R/2.2% O (total 3,116,007; 2.4% of nationwide)

This bucket has a variety of smallish cities, ranging from Spokane and Billings in the west, to Duluth and Appleton in the Midwest, to Binghamton and Troy in the northeast, characterized by being sort-of-industrial but sort-of-service-sector, but definitely very white. It also has a lot of Midwestern small towns, some downscale suburbs like Washington Co. outside Pittsburgh and Johnson Co. outside Indianapolis, and a few odds and ends, like Rutland, considered to be the most blue-collar part of Vermont.

Very low diversity; Low/very low education

VERY LOW DIVERSITY/LOW EDUCATION

Total population: 18,517,274

416 counties; avg. population 44,512

Largest examples: Ocean, NJ (Jersey Shore); Bristol, MA (New Bedford); York, PA

2016: 32.5% C/61.3% T/6.2% O (total 8,700,997; 6.4% of nationwide)

2012: 41.2% O/56.9% R/1.9% O (total 8,338,057; 6.5% of nationwide)

With 416 counties, many of which have four-digit populations, we’re getting down into true rural America here, for the most part. The few large counties here tend to be blue-collar parts of the northeast, like the declining port of New Bedford as well as, further down the list, places like Wilkes-Barre and Erie, Pennsylvania, and Canton, Ohio. Ocean County, New Jersey, isn’t industrial, but it does have a lot of retirees from these other blue-collar places, as well as a lot of struggling tourist- and service-sector workers. But the vast majority of these counties are small, depopulated counties in the Midwest, as you can see from the broad smear of red across the final map.

VERY LOW DIVERSITY/VERY LOW EDUCATION

Total population: 24,811,510

960 counties; avg. population 25,845

Largest examples: Jefferson, MO (St. Louis suburbs); Trumbull, OH (Youngstown suburbs); St. Clair, MI (Port Huron)

2016: 25.2% C/69.6% T/5.1% O (total 10,952,069; 8.0% of nationwide)

2012: 36.3% O/61.7% T/2.0% O (total 10,532,314; 8.2% of nationwide)

Finally, we get to the Very Low diversity/Very Low education bucket, which is also the largest bucket in terms of both population and the sheer number of counties (960, nearly one-third of all counties in the nation). Remarkably, out of those 960 counties, only a couple dozen have a population over 100,000. The vast majority of these counties are in the Midwest, though there are a number of them in the Northwest and even a few outliers in the South (Citrus County, Florida, a retirement destination, and Livingston Parish, Louisiana, a post-Katrina white flight destination outside of Baton Rouge, are two large-ish examples).

If you’re wondering why this bucket wound up significantly larger than any other bucket, that’s because both high levels of whiteness and low levels of education are definitely correlated with ruralness. And if you look at the map, you’ll see that if a county is rural, low population, and in the northern half to the country, it’s very likely it’s in this bucket.

You’ll notice that much of the land mass within Pennsylvania, Ohio, Michigan, Wisconsin, and Iowa falls within this bucket. While those counties represent only a small fraction of those states’ populations, the sharp Clinton dropoff in those counties was enough to narrowly flip those states (and almost enough to flip Minnesota as well). Also noteworthy is the large patch of orange counties in southwest Wisconsin and northeast Iowa. These counties were ones that Obama actually won, but Clinton lost; while, again, that’s a very small part of the overall states, that was instrumental in the losses in those states (and the similar patch of orange in Maine was enough to lose the electoral vote in ME-02).

Blue: high diversity/high education; yellow: high diversity/low education; green: low diversity/high education; red: low diversity/low education

Finally, there’s one more map I’d like to share that tries to combine some of the data from all the previous maps in one place. One map distinguishing all 25 buckets would be ideal, but a map with 25 different color codes is likely to be indecipherable. So, instead, I’ve created a map that only uses four colors for the four quadrants of the chart (which, unfortunately, splits the “average” quintiles down the middle, which makes it a little harder to do a direct comparison). Blue represents the mostly-Dem counties that are above average in both diversity and education; red represents the mostly-GOP counties that are below average in both diversity and education. The yellow counties are above average in diversity and below average in education; the green counties are below average in diversity and above average in education. While the giant swath of red across most of the country looks terrifying, keep in mind that each of the four colors represents about one-quarter of the nation’s population.

Even with all those maps, you still might not be able to get a close-enough look to see which specific bucket your county (or any other county you’re interested in) falls in. To help with that, I’ve put together a Google doc that gives the relevant information for each and every county and county equivalent.