If you’ve spent any time perusing maps of different presidential elections, you’ve probably noticed that, historically, the map would often change wildly from one election to the next. Even in recent history, we’ve seen some presidential maps that seem almost inconceivable right now. Can you imagine a map where the Republican wins California, Illinois, Oregon, and Vermont, but the Democrat wins Alabama, Arkansas, Mississippi, and West Virginia? It may have happened in your lifetime; that describes Jimmy Carter’s victory in 1976.
So when the map stays almost entirely stagnant between two presidential elections, that’s very unusual, historically. The difference between 2008 and 2012 was, in fact, the least change that’s ever occurred between two presidential maps, with only two states flipping (Indiana and North Carolina, from blue to red). Starting in the ‘00s, we’ve entered an era of much greater polarization and negative partisanship, where the parties are more ideologically distinct, there are fewer swing voters, and most states’ votes are extremely predictable, leaving only a dozen or so swing states.
It’s distinctly possible that we could see even less change than that between 2012 and 2016, though. For much of the year if you looked at state-level polling, it looked like only one state was on track to flip (North Carolina, back to blue). In fact, that’s still the case if you look at Huffington Post Pollster’s state-by-state averages; if you take the two states that they currently see as tied (Iowa and Ohio) and award them to Hillary Clinton, North Carolina is the only state where the party that lost in 2012 is currently leading.
However, the Daily Kos Elections model is a little more pessimistic, methodologically, than the HuffPo averages. Because we add in the effect of the “fundamentals” to account for the uncertainty of how the remaining month and a half will go (and because the fundamentals act as a negative, given the difficulty of what Clinton is trying to do: run for a third term for the party in power, against the backdrop of so-so economic numbers), we see her odds of winning North Carolina (despite her small lead in the polling average) as slightly below 50-50 (closer to 40 percent). And our model sees two other states likelier to flip than North Carolina, unfortunately both in the wrong direction: Iowa and Ohio. Those are the two states that have really gone the wrong way for Clinton in the polls in the last few weeks, with multiple polls in each state giving Donald Trump a lead, and that, more than anything else, has had a negative impact on Clinton’s overall odds in our model (especially in Ohio, thanks to its 18 electoral votes).
The problems in Iowa and Ohio may seem odd, given that those two states weren’t the states with the narrowest margins in 2012; those were Florida and North Carolina. It becomes clearer, though, once you look at some of the underlying demographics. Over the fold, I’ll show you some tables, which may provide an “aha!” moment about these two laggards.
Let’s start by describing what has changed in terms of the demographic composition of the parties in the last decade. This chart from Pew Research that was part of a study released last week on party identification is perhaps the clearest illustration you’ll see (scroll down to the third chart in the article, “GOP gains in affiliation...”). Gains in self-described Republican party ID came almost entirely within the white portion of the electorate. In other words, whites in 2008 broke 46 Republican, 44 Democratic; in 2016, they broke 54 Republican, 39 Democratic. (Blacks and Hispanics didn’t change in that period, and Asians actually became more Democratic in that period.)
But if you break whites down further along other demographic lines, you can see that those gains came almost entirely in the 65 or older cohort, and among those who aren’t a college graduate. Among whites who are college graduates, they went from 46 Republican, 46 Democratic in 2008 to 48 Democratic, 47 Republican in 2016. In that same period, whites with a high school diploma or less went from 45 Republican, 44 Democratic in 2008 to 59 Republican, 33 Democratic in 2016!
In other words, you’d expect that the places that have higher concentrations of white voters, especially ones who didn’t complete college, are ones that are moving away from the Democrats. By contrast, places that are more diverse, and have more college-educated people, are the ones moving toward the Democrats. Luckily, that’s easy information to find, thanks to the Census Bureau’s American Community Survey. (Here, I’ll be using their 2014 1-year sample.)
First, though, let’s have some context about how the 2012 battleground compares with this year’s, by showing the rank order among the broadly defined swing states. Let’s limit ourselves to the 15 states where the 2012 margin was within single digits … which, not coincidentally, are the same 15 states that people are most concerned about this year. The “2016” number is the margin according to Huffington Post Pollster’s averages. The “PVI” number is “partisan voting index,” which is political prognosticator shorthand for the disparity between the national average margin and a state or district’s margin, which is helpful for seeing how a state performed across multiple elections. (If you’re wondering why I’m not using our own numbers from the Daily Kos Elections model, for starters, we don’t use national polls, so we don’t have a national poll average baseline to compare against the states.)
2012 results (national margin Obama +3.9)
1. Michigan: Obama +9.5 (PVI D+5.6)
2. Minnesota: Obama +7.7 (D+3.8)
3. Wisconsin: Obama +6.9 (D+3.1)
4. Nevada: Obama +6.7 (D+2.8)
5. Iowa: Obama +5.8 (D+2.0)
6. New Hampshire: Obama +5.6 (D+1.7)
7. Pennsylvania: Obama +5.4 (D+1.5)
8. Colorado: Obama +5.4 (D+1.5)
9. Virginia: Obama +3.9 (D+0.02)
10. Ohio: Obama +3.0 (R+0.9)
11. Florida: Obama +0.9 (R+3.0)
12. North Carolina: Romney +2.0 (R+5.9)
13. Georgia: Romney +7.8 (R+11.7)
14. Arizona: Romney +9.0 (R+12.9)
15. Missouri: Romney +9.4 (R+13.2)
Compare that to this year:
2016 poll averages (national margin Clinton +4.1)
1. Minnesota: Clinton +9.0 (PVI D+4.9)
2. Virginia: Clinton +7.2 (D+3.1)
3. Pennsylvania: Clinton +6.7 (D+2.6)
4. Wisconsin: Clinton +6.3 (D+2.2)
5. Colorado: Clinton +6.0 (D+1.9)
6. Michigan: Clinton +5.6 (D+1.5)
7. New Hampshire: Clinton +5.2 (D+1.1)
8. Florida: Clinton +2.7 (R+1.4)
9. North Carolina: Clinton +1.9 (R+2.2)
10. Nevada: Clinton +0.7 (R+3.4)
11. Ohio: Clinton +0.1 (R+4.0)
12. Iowa: Trump +0.1 (R+4.2)
13. Arizona: Trump +1.6 (R+5.7)
14. Georgia: Trump +1.9 (R+6.0)
15. Missouri: Trump +5.2 (R+9.3)
So, you can see that Iowa has the biggest drop, from #5 down to #12. Nevada also falls a lot, from #4 to #10. Ohio doesn’t fall a lot in terms of its rank order, though the relative drop in PVI is still pretty notable, from R+1 to R+4 (though, again, not as big as Iowa, which went from D+2 to R+4). The biggest leap goes to Virginia, which goes from #9 up to #2, with a PVI gain from even to D+3.
Now let’s think back to the demographic patterns we saw. What happens when we take those same 15 swing states, and rank-order them according to how white they are? (Or more specifically, what percentage of the population is non-Hispanic white. The Census Bureau asks separately about Hispanic status from race, so a respondent could be both “white” and “Hispanic.”)
2014 percent non-Hispanic white (national average 61.9)
1. New Hampshire: 91.2
2. Iowa: 87.1
3. Wisconsin: 82.2
4. Minnesota: 81.3
5. Ohio: 80.0
6. Missouri: 79.9
7. Pennsylvania: 77.7
8. Michigan: 75.7
9. Colorado: 68.8
10. North Carolina: 64.0
11. Virginia: 63.0
12. Arizona: 56.1
13. Florida: 55.6
14. Georgia: 54.2
15. Nevada: 51.3
That tells us part of the story, already. You can see that Iowa and Ohio are near the top of the list of the whitest swing states. However, there are also some inconsistencies. You might be thinking, well, New Hampshire and Minnesota are pretty white, but they aren’t underperforming this year. And look at Nevada: it’s the most diverse of all the swing states, and yet it’s fallen way down the totem pole this year.
To get a fuller picture, let’s look at another demographic angle: education. Here’s the rank order of the percentage of residents 25 years or older, who have a bachelor’s degree or more.
2014 percent with bachelor’s degree or more (national average 30.1)
1. Colorado: 38.3
2. Virginia: 36.7
3. New Hampshire: 35.0
4. Minnesota: 34.3
5. Georgia: 29.1
6. Pennsylvania: 29.0
7. North Carolina: 28.7
8. Wisconsin: 28.4
9. Iowa: 27.7
10. Arizona: 27.6
11. Missouri: 27.5
12. Michigan: 27.4
13. Florida: 27.3
14. Ohio: 26.6
15. Nevada: 23.1
Adding education adds a lot more clarity. If you were asked “which states are overperforming the most this year,” you’d instinctively say Colorado and Virginia, which indeed have shot up the totem pole. And voilà: Colorado and Virginia have the highest percentages of college-educated residents. That’s followed by New Hampshire, which unlike Colorado and Virginia is very white, but is also holding steady as part of the Democrat firewall thanks to its educated electorate. Contrast that with Iowa and Ohio, which are noticeably less college-educated than average, and also with Michigan, which is still putting up pretty solid numbers for Clinton this year but has fallen several slots down the list. And, finally, look at Nevada: it’s the least-college educated of all the swing states, which may help explain why it’s lagging this year. (Remember Nevada is the state where Donald Trump proudly announced after his caucus win, “I love the poorly educated!”)
Of course, you might reasonably think, “OK, Nevada is the most diverse swing state; it has a lot of Hispanic and Asian immigrants, many of whom haven’t had the chance to go to college. They aren’t voting for Trump anyway, so we need a different theory.” Well, luckily, if you dig deep on their site, the Census Bureau also offers a breakout of the percentage of only non-Hispanic whites who have or haven’t gone to college! (The table has the rather cumbersome name of “sex by educational attainment for the population 25 years and over (white alone, not Hispanic or Latino).”)
2014 percent of non-Hispanic whites with bachelor’s degree or more (national average 33.6)
1. Colorado: 44.4
2. Virginia: 40.3
3. Minnesota: 35.7
4. New Hampshire: 35.0
5. Arizona: 33.6
6. Georgia: 33.1
7. North Carolina: 32.2
8. Florida: 30.9
9. Pennsylvania: 30.5
10. Wisconsin: 29.9
11. Michigan: 28.4
12. Missouri: 28.4
13. Iowa: 28.0
14. Nevada: 27.7
15. Ohio: 27.6
When you layer race and education on top of each other, the pattern really leaps out at you. The states with above average levels of college-educated white voters are the ones that are moving up the chart: either the swing states that are becoming fully blue (Colorado, Virginia, New Hampshire), or the light-red states that are becoming swingy (the “New South” states of Georgia and North Carolina, along with Arizona).
And look who’s at the bottom of the list. It’s the states that have, relatively speaking, gotten worse for the Democrats this year: Iowa, Nevada, and Ohio. (And to a lesser extent, Michigan and Wisconsin, which still look like they’re staying in the Dem column without much trouble, but are still getting lapped by the Virginias and Colorados of the world … and Missouri, which looks like it’s going to stay out of reach again this year.)
There are, of course, nuances that are missing from the relationship that I’ve shown; for instance, Ohio is probably more receptive to Trump’s platitudes about trade and manufacturing than voters in Colorado or Virginia, where there’s more knowledge sector/creative class employment. Of course, that’s already baked into the education numbers, though, so it’s all rather circular (you need a college education to get those same information economy jobs that are prevalent in Colorado and Virginia).
Turning back to the model itself, the recent declines that we’ve seen specific to Ohio and Iowa are what’s driving the continued erosion in Clinton’s overall odds, which are currently at 65 percent. One consideration that I didn’t address last Thursday, when writing the day after two especially bad Ohio polls came out (and the overall odds were still 73 percent at that point), was that we have some anti-volatility tweaks built into the model, so that polls are downweighted for the day immediately after their release, so they don’t land with a huge thud at first and cause double-digit one-day swings.
So, at this point, we’re seeing a full impact of those Ohio polls, as well as the poll from Emerson College Polling Society in Colorado that gave Donald Trump a lead (which, of course, is very inconsistent with everything else we’ve seen in Colorado, but still needs to go on the pile). The good news is that, by the same logic, today’s numbers don’t fully reflect the Muhlenberg poll in Pennsylvania from over the weekend that gave Clinton a 9-point lead; as that phases in, it should further bolster her already-robust odds in the Keystone State and thus overall. (Remember that the easiest route to 270 is winning Colorado, Michigan, New Hampshire, Pennsylvania, Virginia, and Wisconsin, all of which still show substantial Clinton leads in the aggregate.)
On the other hand, you might notice that Democratic odds in the Senate model have improved a bit, up to 49 percent. That’s also the result, in part, of Emerson’s erratic polling; they were the first pollster to find Jason Kander leading Republican incumbent Roy Blunt in the Missouri Senate race, improving Kander’s overall odds significantly there. While Blunt is still favored, the growth of the Missouri path gives the Dems one more route to 50. (And the DSCC clearly understands that, as seen in their recent moving of money from the Ohio and Florida races to Missouri and North Carolina.)