With Republican Sen.-elect John Kennedy’s triumph in the Louisiana runoff last weekend, victories by two other Republicans in Louisiana House races, and Gov. Pat McCrory’s (R) concession last week to Gov.-elect Roy Cooper (D) in North Carolina, the winners of 2016’s House, Senate, and gubernatorial races are now set. This allows us to do a little housekeeping. Kennedy’s win confirms that this is the first cycle in the history of popular Senate elections that every state that held a Senate election in a presidential cycle voted for the same party for both president and for Senate (34 for 34 this year). Also, finalizing these results permits us to give a final assessment of our down-ballot Crystal Ball projections for 2016: We picked 32 of 34 Senate races correctly, along with 10 of 12 gubernatorial races and 428/435 House races.

Looking over the down-ballot outcome, there’s one inescapable conclusion in a year that was defined by a political outsider, Donald Trump, winning the presidency: It was still a really good year to run as an incumbent in 2016, all things considered.

This election cycle, 393 of 435 House representatives, 29 of 34 senators, and five of 12 governors sought reelection (several of the governors were prohibited from seeking another term). Of those, 380 of 393 House members (97%), 27 of 29 senators (93%), and four of five governors (80%) won another term. These members of Congress and governors not only won renomination, but also won in November.

Those reelection rates are all a little bit better than the already impressive post-World War II averages, as shown in Table 1.

Table 1: Incumbent reelection rates in 2016 compared with post-World War II historical averages

Sources: Crystal Ball research; Vital Statistics on Congress

This year’s incumbent reelection rates could easily have been even higher. The aforementioned Pat McCrory only lost his reelection bid by two-tenths of a percentage point, and outgoing Sen. Kelly Ayotte (R-NH) only lost by 0.14 points to Sen.-elect Maggie Hassan (D), who did not seek reelection as governor of the Granite State. Additionally, mid-decade, court-ordered redistricting in Florida, North Carolina, and Virginia all contributed to some incumbent losses. New maps played a big role in the primary defeats of outgoing Reps. Corrine Brown (D, FL-5), Renee Ellmers (R, NC-2), and Randy Forbes (R, VA-4 — he actually ran for reelection in VA-2, though, because of redistricting), and also contributed to the general election losses of outgoing Reps. David Jolly (R, FL-13) and John Mica (R, FL-7). Had redistricting not occurred, the small number of losing incumbents (13) almost certainly would have been smaller.

That’s not to say there isn’t a good deal of turnover in these high-level offices. Most states have term limits for governor, and because of incumbent defeats, retirements, and other factors, about an eighth of the U.S. House will be new in 2016 (and several more new members will be elected in special elections next year to replace some House members who are likely to be confirmed for positions in the Trump Administration). Seven of 100 senators will be new members come January, and that number eventually will rise if Sen. Jeff Sessions (R-AL) is confirmed as attorney general and replaced by a gubernatorial appointee or if other members, such as Sen. Heidi Heitkamp (D-ND), are nominated and confirmed for a Cabinet post.

However, despite the fact that there is some churn in Congress and in the statehouses, the reality is that if an incumbent is on the ballot, he or she typically has good odds of winning. That’s been true for much of recent American history, and it was still true in 2016.

Speaking at the 2004 Democratic National Convention, Barack Obama, then a candidate for the U.S. Senate, famously declared that “there’s not a liberal America and a conservative America; there’s the United States of America.” Obama then went on to decry political pundits who “like to slice and dice our country into red states and blue states.” The implication of Obama’s speech was that the perception of geographic sorting of the country into reliably Democratic and Republican areas was not based in fact, but was instead a false narrative imposed by the media.

Obama’s rhetoric of unity and homogeneity across party lines notwithstanding, there is substantial evidence that cultural and lifestyle preferences are strongly related to political tastes. Political scientists have demonstrated that political ideology and party identification are predictive of choices in areas as varied as mate selection, media consumption, cleanliness, office décor, music tastes, and housing decisions. If such relationships between political and lifestyle preferences are strong enough, and individuals are sufficiently willing and able to “vote with their feet” and move to areas that are a better match for their tastes, the implication is clear: We should expect to see the emergence over time of a geographically divided electorate of exactly the kind that Obama’s famous speech in 2004 sought to deny.

Perhaps the most famous argument in support of such a geographically divided electorate can be found in Bill Bishop’s The Big Sort. In this account, Bishop points out that most Americans live in neighborhoods that are dominated by one political party and that this divide has been growing over time. The result has been a “fundamental kind of self-perpetuating, self-reinforcing social division” of Americans along geographic and partisan lines.

While an argument like Bishop’s is persuasive, there has been very little empirical examination of these claims. To understand both whether partisans sort into geographic locales and the degree to which this sorting takes place, Greg Martin and I analyzed two similar sources of data: The first contains information on the moving patterns of the nearly 1.1 million people who moved from one location to another within the state of Florida between 2008 and 2010; the second contains information on 50,000 individuals who moved from one state to another between 2005 and 2016. In both sets of data, we analyzed whether Democrats and Republicans have differential preferences over three distinct characteristics of a geographic locale: the log population density; the “walkability” of a neighborhood, measured by its closeness to restaurants, shops, offices, and parks; and, finally, the Republican share of the two-party vote. While the results we find hold across both of our samples, I only present results from our dataset of Florida movers here.

Given what we know about Democratic and Republican partisans, there are clear expectations about the sorts of patterns we should see if a “big sort” is, indeed, occurring: Democrats (Republicans) should prefer more (less) dense locales, more (less) walkable areas, and areas with a lower (higher) Republican share of the two-party vote. What we find is that Democrats and Republicans do prefer to live in areas with more co-partisans. For instance, registered Democrats moving from one location to another in Florida move, on average, to locations about 5 log points denser than observably comparable registered Republicans. Moreover, registered Democrats move to areas that, on average, have approximately 2% lower Republican vote shares than observably comparable Republicans. These results are shown graphically in Figure 1.

Figure 1: Regression coefficients for Florida movers, 2008-2010

While these results suggest that there is a degree of partisan sorting along geographic lines taking place within the American electorate, the more important factors in determining patterns of migration are nonpolitical attributes of a location. Indeed, when we factor in characteristics of a neighborhood such as the racial, gender, and age demographics; income deciles; education levels; levels of home ownership in the area; and qualities of a house, such as the median number of rooms and the year it was built, the influence of partisanship in determining where an individual moves diminishes greatly. Americans, then, prefer to live near co-partisans, but this preference is a much lower priority than more practical considerations.

Given that our data suggest that partisanship is a very small factor in determining where an individual chooses to live, one interesting question is how long it would take to achieve a geographically polarized electorate like the kind we see in contemporary U.S. politics. Therefore, in order to answer this question, we simulated 20 years’ worth of moving in our data set of Florida movers. This simulation process consists of a few steps: First, we estimate the probability that every voter registered in Florida between 2008 and 2010 moves between 2008 and 2010; next, those individuals who are predicted to move relocate to a new precinct that better fits their own partisan preferences; after this, we recompute the Republican presidential vote and registration shares in every precinct once moving is complete. We repeat this process for a period of 20 years or, equivalently, 10 electoral cycles. After simulated each cycle, we record and plot the proportion of the Republican voter share and the Republican share of two-party registrations in each of Florida’s 25 congressional districts (based on the 2008 starting point; Florida has 27 districts following the 2010 reapportionment). These simulated changes are shown in Figures 2 and 3.

Note that we treat individuals’ presidential votes and registration choices as fixed; the only thing that may vary over time is their residential location. The statewide population of voters and hence statewide Republican shares are fixed throughout the course of the simulation. All that may potentially change is the composition of lower-level geographic units such as precincts or state house or congressional districts.

Figure 2: Simulated change in Republican presidential vote share by congressional district, 2008-2028

Figure 2: Simulated change in Republican share of two-party voter registration by congressional district, 2008-2028

As can be seen in Figures 2 and 3, both the proportion of the Republican voter share and the Republican two-party registration share increases very little over time. In fact, any change that does occur tends to produce more competitive congressional districts rather than more districts that are dominated by one party. This is because, as we have suggested, Americans are more concerned with things like finding affordable homes in areas with good schools than living near other people who share their partisan affiliation. Given the estimates we have found in our analyses, it would take an extraordinarily long time — if ever — for geographic polarization along partisan lines to occur. This claim is shown clearly in Figures 4 and 5. Figure 4 shows the distribution of simulated 2008 Republican presidential vote share at the precinct level in each of the 10 simulation cycles. Figure 5 shows the distribution of two-party Republican registration shares at the precinct level over this same simulated time period. If geographic sorting was happening, we should see an increasingly bimodal distribution on both of these metrics. Instead, what we see is the emergence of more competitive precincts and the decline of more lopsided precincts.

Figure 4: Distribution of simulated 2008 Republican presidential vote share at the precinct level, 2008-2028

Figure 5: Distribution of two-party Republican voter registration shares at the precinct level, 2008-2028

The results shown above suggest that partisan geographic sorting, in isolation, has a homogenizing — rather than polarizing — effect on the distribution of political preferences. What, then, explains the apparently stable degree of polarization in the current observed distribution of preferences? One possibility is that individuals’ political preferences are influenced by their social and physical environment. When a rural voter moves to an urban location or vice versa, his or her political preferences may change to fall in line with those of his or her new peers and neighbors, just as other cultural and lifestyle preferences are likely to change following such a move. Our simulation exercise explicitly rules this possibility out: individuals’ political preferences are assumed to be completely fixed. Under this assumption of fixed preferences, residential mobility tends to produce politically homogeneous rather than segregated geographic distributions. The fact that we do not observe such a homogeneous pattern, then, suggests that the reverse causal channel (location changing preferences, rather than preferences changing location) has some bite.

In order to examine this possibility, we examine changes in individual voters’ partisan registration over time. Specifically, we ask if voters who moved to new locations that were very different from their old locations on the three dimensions discussed above (population density, walkability, and GOP vote share) subsequently changed their registration status in a direction matching the change in their environment. Put more succinctly, do Democrats become more Republican over time when they move to more Republican areas (and vice versa)? Our analysis suggests that this process is, indeed, happening. When Republicans move to areas that are more densely populated, have higher degrees of walkability, and are more Democratic, these individuals — over time — shift their party affiliation from Republican to Democrat (or Independent). Conversely, Democrats who move to areas that are less densely populated, have lower degrees of walkability, and are more Republican, tend to become Republicans (or Independents).

The American electorate is geographically divided. Partisan affiliations correlate with tastes for housing characteristics, with Democrats clustered in urban cores and Republicans spread out across suburban and rural areas. However, our analysis suggests that this spatial distribution of partisans is not the result of any form of systematic partisan sorting. Individuals both move frequently enough and care more about “practical” considerations than political ones when deciding where they want to move. Thus, we do not believe that a “big sort” is occurring. We do, however, find evidence that where one chooses to live is important inasmuch as one’s peers have the ability to influence one’s own partisan affiliation. Whether this process is happening due to social influence, a desire to not stand out, or something else is a question that remains to be answered.

To read more, visit http://polisci.emory.edu/faculty/gjmart2/papers/partisan_sorting_density.pdf  or http://www.stevenwwebster.com/research/partisan_sorting_density.pdf.

The Crystal Ball is coming out a day early this week because we wanted to invite our readers to watch the University of Virginia Center for Politics’ 18th annual American Democracy Conference, which is going on today from 9 a.m. to 3:30 p.m. at the Ronald Reagan Building & International Trade Center in Washington D.C. Registration is full but members of the media interested in covering the conference are welcome to attend, and we will be streaming the conference live throughout the day at https://livestream.com/tavco/ADC2016.

Our featured speakers are Kellyanne Conway, who managed Donald Trump’s successful presidential campaign, at 9 a.m., and Sen. Susan Collins (R-ME) at 12:30 p.m. Additionally, Center for Politics Director Larry J. Sabato will interview Khizr Khan, the Gold Star father who spoke at the Democratic National Convention, at 1:30 p.m. There will also be three panels discussing various aspects of the 2016 campaign.

Here’s the full program:

9:00 a.m.

Welcoming Remarks by Larry J. Sabato, Director, UVA Center for Politics

9:05 a.m.

Keynote Speaker: Kellyanne Conway, campaign manager for President-elect Donald J. Trump’s campaign

9:45 a.m. – 11 a.m.

Panel I – Politics and the Fourth Estate

Moderator: Geoffrey Skelley, Associate Editor of Sabato’s Crystal Ball

Molly Ball, National Politics Staff Writer for The Atlantic

Jamelle Bouie, Chief Political Correspondent for Slate

Karen Tumulty, National Political Correspondent for The Washington Post

Carlos Watson, Editor in Chief, OZY.com

11 a.m. – 12:15 p.m.

Panel II – Divided States of America

Moderator: Kyle Kondik, Managing Editor of Sabato’s Crystal Ball

Alan Abramowitz, Sabato’s Crystal Ball senior columnist and Emory University professor

Glen Bolger, Republican pollster and co-founder of Public Opinion Strategies

Angela Rye, Principal and CEO of IMPACT Strategies

David Byler, Elections Analyst for RealClearPolitics

12:15 p.m. – 12:30 p.m.

Lunch

12:30 p.m. – 1:15 pm.

Keynote Speaker: Sen. Susan Collins (R-ME)

1:15 p.m. – 1:30 p.m. 

Presentation by Alan Abramowitz, Emory University professor

1:30 p.m. – 2:15 p.m. 

Larry Sabato interviews Khizr Khan, Gold Star father who spoke at the Democratic National Convention

2:15 p.m. – 3:30 p.m. 

Panel III – Political Perspectives on 2016 and Beyond

Moderator:  Carol Costello, CNN

Perry Bacon Jr., Senior Political Reporter, NBC News

Fred Barnes, Executive Editor of The Weekly Standard

Joe Borelli, Council member, 51st District and Minority Whip of New York City Council

Maria Cardona, Democratic strategist and principal at Dewey Square Group

Editor’s Note: The Crystal Ball is taking the week off for Thanksgiving next week, but we’ll be back with another edition on Thursday, Dec. 1.

Now that we’ve had a week to digest the results of the 2016 election, here are some observations about what happened and what the results might tell us about the future:

1. Electoral map tilts to the GOP

In close elections, the Electoral College will probably continue to tilt to the GOP. Twice in 16 years, we’ve had a “misfire,” where the popular vote went to one major-party candidate while the other candidate secured a majority of the electoral vote. This is because Democrats secure large, sometimes enormous, majorities in mega-states such as California, New York, and Illinois, while Republicans have just Texas, where Donald Trump’s margin of victory was nearly 450,000 votes fewer than Mitt Romney’s. (You should never join “just” with Texas, but we trust you’ll see what we mean.) Other sizable states, such as Florida, North Carolina, and Virginia, are closely divided and add only small pluralities to the candidate that wins them.

While the cumulative popular vote means nothing under the Constitution, it is not a good thing for a president to have lost it. The powers of the office are the same for every chief executive, yet it is almost impossible for a president to claim a mandate when many more people voted for the opponent. Democratic candidate Al Gore secured a national plurality of 547,000 votes in 2000; as of Wednesday afternoon, Hillary Clinton has 1.2 million more votes than Donald Trump, and that gap is very likely to grow.

Most Americans have long favored abolition of the Electoral College, designed for the 1790s rather than the 21st century. However, no one expects to see the Electoral College go the way of the horse and buggy anytime soon. It is also fair to note that if there were no Electoral College, the campaigns would have been run differently, so we can’t automatically assume that Hillary Clinton would have beaten Donald Trump under a popular vote system.

2. The map gets more competitive

Because the 2016 contest was far closer than either 2008 and 2012, the number of states decided by five points or less increased from four in 2012 (Florida, North Carolina, Ohio, and Virginia) to 11 in 2016 (shown in the map below). But the nation’s polarization is still obvious, and only six states flipped from Democratic to Republican (Florida, Iowa, Ohio, Pennsylvania, Wisconsin and probably Michigan) — and none in the opposite direction. At least we can say that the playing field was more broadly competitive, and especially so until the final GOP consolidation behind Trump in late October and early November. For a while in the fall, Arizona, Georgia, Utah, and a few other states appeared to be tightly contested, and even on Election Day Arizona and Georgia ended up being less Republican than Iowa and Ohio. We may be at a time of transition in the Electoral College where the whiter Midwest gets redder and the more diverse Sun Belt gets bluer.

Map 1: Electoral College results and margins of victory

Source: Data from Dave Leip’s Atlas of U.S. Presidential Elections

3. Projections miss the mark

With credit to 270toWin.com, here is the final Election Day forecast for a range of sites and news organizations. No one was even close to the final Electoral College results. The Crystal Ball actually had the largest allocation of electors to Trump (216) but we were in a three-way tie for the second-most electors assigned to Clinton (322). This will be a year pundits and predictors will want to forget, yet none of us should. There are many lessons to learn, and we all need to do much, much better in the future.

Table 1: 2016 Electoral College projections as compiled by 270toWin.com

4. Many polls were off, but some provided clues in the key states days before the election

By the time all the votes are counted, the RealClearPolitics average of national polls that showed Clinton winning the national popular vote by about three points probably won’t be all that far off — Clinton may win it by a point or two or somewhere in between. But many of the state-level polls missed the mark, and missed badly, particularly in the Rust Belt. Hardly any polls showed Trump leading in several of these states the whole election, and yet Trump’s one-point wins in Pennsylvania and Wisconsin and even narrower lead in Michigan, which the Associated Press has yet to officially call, ended up being the difference in the race: had Clinton carried these three states, she would have won with 278 electoral votes (as it stands now, if Trump carries Michigan he’ll end up with 306).

But there were some signs. A Republican pollster, the Trafalgar Group, released polls of Michigan and Pennsylvania showing Trump with tiny leads. Another Republican pollster, Harper, showed the race tied in Pennsylvania. It’s hard to believe partisan polls before Election Day, but they are sometimes correct when a contest is breaking their party’s way late, as appears to have been the case in 2016.

Additionally, Ann Selzer, the respected nonpartisan Des Moines Register pollster, had Trump up seven points in Iowa in a poll released the Saturday before the election. Trump ended up winning Iowa by nine points, and it looked like a Trump state for much of the fall, but Selzer’s poll could or should have been a warning about other Midwest states. If Iowa were truly going for Trump by that large margin, it probably meant that states like Michigan and Wisconsin — states filled with many of the same working-class white voters who would power Trump to victories across the Heartland — were going to be very close, which they were.

5. Michigan: Poor urban turnout tells the tale

Despite doing significantly worse outside of Wayne County (Detroit) than Barack Obama, if Hillary Clinton had replicated the margin the county provided any of the last three Democratic presidential nominees (342,000, 441,000, and 382,000 in 2004, 2008, and 2012, respectively), she would have carried the state. As of the most recent count, Clinton is losing Michigan by about 13,000 votes, with only about a 289,000-vote margin in Wayne, about 53,000 votes worse than John Kerry’s.

6. But not necessarily in Pennsylvania

A vital part of the Democratic formula in Pennsylvania is getting a huge margin out of Philadelphia County, particularly in recent years when some typically reliable Democratic counties in Appalachian western Pennsylvania have deserted the party. Unlike in Detroit, where Clinton’s sagging margin was likely decisive, Clinton did acceptably in Philadelphia. Her margin in Philadelphia was about 457,000 votes, down from Obama’s 479,000 and 492,000 in 2008 and 2012, respectively, but a decent win, and markedly better than John Kerry’s 412,000-vote margin in 2004 (Kerry won Pennsylvania).

Clinton also outperformed Obama in Allegheny County (Pittsburgh), adding about 15,000 votes to Obama’s 2012 margin. So why did she lose the state by about 66,000 votes? Because outside of Greater Philadelphia and Pittsburgh and its closest suburbs, she lost by huge margins.

Leaving aside Allegheny and Philadelphia counties, Barack Obama lost the rest of Pennsylvania by about 273,000 votes. Clinton lost the rest of the state by…wait for it…629,000.

A great example of Clinton’s struggles outside of Pittsburgh and Philadelphia is Lackawanna County, a white, working class enclave that contains Scranton. Obama won the county by slightly more than 25 points in each of his elections, but Clinton won it by only three percentage points. Clinton also lost Erie County in the northwestern corner of the state — it swung from Obama +16 to Trump +2. There are similar eye-popping swings across the state. Brandon Finnigan of the election results-reporting site Decision Desk HQ has argued for years that Republicans had a path in Pennsylvania, and he was proven correct thanks to the giant shift that the Trump candidacy effected in the Keystone State.

7. Or Wisconsin

The Wisconsin story is similar. Clinton lost the state by about 24,000 votes, but her combined margins from the state’s two Democratic powerhouse counties — Milwaukee and Dane (Madison) counties — were almost identical to Obama’s margins in 2012. Clinton’s margin dipped a bit in Milwaukee from Obama but she did better in Dane than Obama, and she only lost about 800 net votes from Obama in those two counties combined, effectively a wash. Outside Milwaukee and Dane, Obama lost Wisconsin by 97,000 votes. Clinton lost outside of Milwaukee and Dane by 333,000 votes.

To win Pennsylvania and Wisconsin in 2020, Democrats are going to have to work to reverse some of the transformation outside of Philadelphia, Pittsburgh, Milwaukee, and Madison. Both states have what should be highly competitive Senate and gubernatorial races in 2018, and the performance of the small cities and rural counties that swung hard from Obama to Trump will be interesting: Will they stay Republican without Trump on the ballot, or will they revert to their Democratic roots?

8. Ohio: Bellwether no more?

Ohio voted for the presidential winner for the 29th time in 31 elections. That’s the best of any state in that timeframe, which the Crystal Ball’s Kyle Kondik explored in his book on Ohio, The Bellwether. Additionally, Ohio built its lead over its other competitors for the “bellwether” title in this election because the states with the records closest to Ohio’s — New Mexico, Illinois, and Nevada — all voted for the Electoral College loser (Clinton) in this election.

However, Ohio may be trending Republican and away from being the nation’s top bellwether. Trump won the state by 8.5 points at the most recent count, while Clinton was winning by at least one point nationally. Ohio will be further from the national average in this election than it has been in any election since before the New Deal. Part of being a bellwether state is not just voting for the winner, but also voting close to the national average — something that Ohio has done consistently for decades but did not do in this election.

As we observed before Nov. 8, if the changes Trump made to the electorate this year are durable, Ohio may be the new Missouri: a longstanding bellwether state that in recent years has clearly trended Republican. (Speaking of Missouri, which voted for the presidential loser only once between 1904 and 2004, the state had its biggest Republican lean relative to the national result in this election since the Civil War.)

9. Why Colorado, Nevada, and Virginia didn’t flip

While the Rust Belt largely abandoned the Democrats in 2016, albeit by very small margins in some of the aforementioned states, states with newer Democratic leans like Colorado, Nevada, and Virginia held fast for Clinton. There are two demographic factors that likely help explain why this occurred, particularly in comparison to the three Rust Belt States that crucially flipped for Trump: nonwhites, especially Latinos in Colorado and Nevada, and college-educated white voters. Helpfully for Trump, the exit polls found that white voters without a college degree made up pluralities of the electorates in Michigan and Wisconsin and were about even with white college-educated voters in Pennsylvania. Although Colorado had a smaller nonwhite share than Michigan, its electorate had a large plurality of white college-educated voters. Virginia also had a plurality of whites with college degrees and a sizable nonwhite share of the electorate. And Nevada had a plurality nonwhite electorate, the largest nonwhite share of the states in question, helping Clinton overcome the slightly larger share of voters who were white without a college degree in the Silver State.

As of the most recent count, Clinton won Colorado by a little under five points, Virginia by a little over five, and Nevada by about 2.5 points.

Table 2: Exit poll data for shares of electorates in Rust Belt trio and Sun Belt trio

Source: CNN

10. The increasing urban-rural divide

The political split between more urban and more rural areas only grew in 2016. In 2012 Mitt Romney carried just four of today’s 50 most populous counties. Donald Trump carried the same number of localities, but with four changes. Romney carried Salt Lake County, UT — not shocking given his Mormon faith — and Orange County, CA. Trump lost both places — notably, Orange County had not voted for a Democrat for president since Franklin Roosevelt in 1936. But unlike Romney, Trump carried battleground Pinellas County, FL along the important I-4 corridor in the Sunshine State, as well as Suffolk County, NY, at the eastern edge of Long Island.

Table 3: 2016 and 2012 presidential vote in the 50 most populous counties as of July 1, 2015

Note: 2016 election data as of Nov. 16, 2016

Source: Dave Leip’s Atlas of U.S. Presidential Elections, Associated Press (IL, MA, ME, MS)

Among the largest localities in the country, Trump did worse overall than Romney while making up ground in places with smaller populations. Clinton performed about 0.5 points better (California still has many votes left to be counted) in the 50 counties in Table 3 than Obama did in 2012, but Trump performed about four points worse that Romney in the same group. However, in counties around the country with populations below 200,000 people, Trump improved about 4.5 points on Romney’s 2012 performance while Clinton did 7.5 points worse than Obama.

11. Appalachia continues to trend away from the Democrats

Democrats have been losing ground in Appalachia, a working-class and overwhelmingly white region that stretches from western New York to northern Alabama and Mississippi, for years. The most obvious early sign was George W. Bush’s victory in West Virginia in 2000. In 1988, West Virginia was just one of a handful of states to vote for Democrat Michael Dukakis, but by Bush’s first election it had become more Republican than the nation, and its GOP lean has only grown. Trump’s pro-coal, anti-internationalist, “America First” campaign seemed optimally suited for Appalachia, and the results bore that out: Trump won West Virginia, the only state which is fully contained in Appalachia, by nearly 42 points (his second-best state margin, behind only GOP stalwart Wyoming).

Clinton won only 6% of all the Appalachian counties, down from the paltry 8% that Obama won in 2012. As is clear from Table 4, Democrats used to do quite well in the region, but now their support in the area is largely confined to places like Allegheny County (Pittsburgh), some majority-minority counties in Alabama and Mississippi, and liberal college enclaves like Athens County, OH (home to Ohio University) and Tompkins County, NY (home to Cornell University and Ithaca College).

Table 4: Selected Democratic presidential performance in Appalachia 1976-2016

Note: Counties included are the 428 counties and independent cities classified as part of Appalachia by the Appalachian Regional Commission. Appalachia, as defined by the commission, covers all of West Virginia and parts of Alabama, Georgia, Kentucky, Maryland, Mississippi, New York, North Carolina, Ohio, Pennsylvania, South Carolina, Tennessee, and Virginia.

Meanwhile, some of the swings in these counties were stunning. In Ohio, Obama lost Pike County, a small, historically Democratic county that in recent years has voted Republican by small margins for president, by a single vote. Clinton lost it by 36 points. Monroe County, another tiny Ohio county that has been so Democratic for so long that the Whigs (the precursor to the Republicans) called it “Dark Monroe” because they had such trouble winning it, was one of only two counties to vote for beleaguered Democratic gubernatorial nominee Ed FitzGerald in 2014 (although it did vote for Romney). Clinton lost it by 46.5 points.

Clinton’s collapse in the region not only helps explain her very poor showing in Ohio, but also her surprising loss in Pennsylvania. Here’s another way of looking at the Keystone State: 52 of the state’s 67 counties are classified as Appalachian, and these counties cast 44% of the state’s total votes in both 2012 and 2016. Obama lost the Appalachian counties by about 175,000 votes. Clinton lost them by about 492,000.

Overall, Trump won Appalachia by an astounding 63%-33%, improving on Romney’s 60%-39% edge in 2012.

12. Democrats more dependent on nonwhite voters

Considering Clinton’s struggles among white working-class voters, seen most clearly in the results in many parts of the Rust Belt, it should be no surprise that the election results suggest that Clinton’s coalition showed greater reliance on nonwhite voters than even Obama’s did. Just consider a simple correlation between Democratic percentage and the nonwhite citizen-voting-age population of U.S. counties. In 2012, that correlation was .48, which is moderate but not remarkably strong. Fast forward to 2016 and the correlation grew to a stronger .61. These data indicate that larger nonwhite populations corresponded to better Democratic performance in 2016 than in 2012. Below is a scatterplot of the 2016 data.

Chart 1: Scatterplot of county nonwhite CVAP share compared to 2016 Democratic vote percentage

While the exit poll suggested that Trump did better among black and Latino voters than Romney, other data have called those findings into question, including this stronger correlation. Furthermore, the reverse is also true for non-Hispanic white areas, with a slightly stronger correlation between the share of non-Hispanic white population and support for Trump compared to Romney (.54 in 2016 versus .46 in 2012). Additionally, counties with large Latino populations saw Republican vote support drop two-to-three points while Democratic vote performance stayed closer to 2012 or even improved by a point or so. (We will return in a future Crystal Ball to the dimensions of Latino backing for Trump. It is very probable that the exit poll overestimated Trump’s strength with Latinos.)

13. The third party fizzle

Ultimately, what happened to Gary Johnson of the Libertarian Party and Jill Stein of the Green Party is what happens to most third-party candidates: They fade at the end. Johnson flirted with 10% national support for much of the race but ended up with just 3.3% of the national vote, while Stein got 1% despite reaching nearly 5% in the RealClearPolitics average in early summer.

Map 2 shows Johnson’s level of support by state. As is common for Libertarians, he did better west of the Mississippi than east of it, and his best state, unsurprisingly, was his own state of New Mexico, where he got 9%.

Map 2: Gary Johnson’s level of support by state

Stein’s best states weren’t surprising either. She did best in the liberal states of Hawaii (where she got nearly 3%) as well as Oregon and Vermont. Stein’s total votes in Michigan and Wisconsin were greater than the margin Clinton lost each state, so had the Stein backers in those states voted as a bloc for Clinton instead, she would have carried each state (although Clinton still would have lost the Electoral College).

Map 3: Jill Stein’s level of support by state

Note: Stein was not listed on the ballot in Georgia, Indiana, Nevada, North Carolina, Oklahoma, and South Dakota.

14. Straight tickets for Senate

The 2016 election saw the largest wave of straight-ticket results for president and Senate since the ratification of the 17th Amendment established popular election of senators across the country in 1913. All 33 states with Senate elections decided on Nov. 8 backed the same party for president and Senate. Although Louisiana’s final outcome won’t be determined until it completes a Dec. 10 runoff election, Republicans are heavily favored to retain it; should state Treasurer John Kennedy (R) win, that would mean 34 of 34 states voted for the same party in presidential and Senate races.

The 100% straight-ticket mark for 2016 is the highest ever, narrowly outdoing 1920’s 97% mark, and it comes on the heels of three presidential cycles (2004-2012) where the straight-ticket percentage had been about 80%. The closest any state came to having a split-ticket result was in New Hampshire, where Sen. Kelly Ayotte (R) lost by about 1,000 votes (less than 0.2 percentage points) to Sen.-elect Maggie Hassan (D), the state’s outgoing governor.

Chart 2: Percentage of straight-ticket vs. split-ticket presidential-Senate outcomes, 1916-2016

Note: Includes all states where both the presidential and Senate races were won by major-party candidates

15. Do the Republicans have new targets in the House?

Before the election, we speculated about the possibility of some white, working-class House districts with Democratic incumbents potentially producing a massive House upset because of their swings to Trump. While there were no upsets, there were a few close calls. Reps. Tim Walz (D, MN-1) and Collin Peterson (D, MN-7) had closer-than-expected races as their districts swung hard to Trump, as did Rep. Matt Cartwright (D, PA-17). As the National Republican Congressional Committee plots its targets for 2018, some of these districts might get special attention. On the flip side, Democrats will be looking to 2018 as an opportunity to make gains in the House, and at this early date, as unlikely as it may seem today, we would not rule out the possibility of Democrats grabbing the roughly 25 net seats they would need to take control. Simply put, midterm elections are often a backlash vote against the president’s party, and with Trump in the White House, history suggests the opportunity is there for the Democrats, even on a national House map that has given them headaches in recent years thanks at least in part to Republican control of redistricting in many swing states following the 2010 census.

16. Democratic losses in the Age of Obama

As we pointed out after the 2014 midterm election, the Democratic bench has taken an unprecedented hit during President Obama’s time in office. The numbers have worsened slightly following Obama’s final election as a part of the political environment. With most 2016 results in (adding projections for some uncalled races based on who is ahead at this point), the damage is as follows: a net loss of 13 governorships, 11 Senate seats[*], 63 House seats, 949 seats in state legislatures, and 29 state legislative chambers. Some other modern presidents lost more governorships, Senate seats, and state legislative chambers, but none has lost more net House seats and — especially — state legislative seats. Having lost close to 1,000 (!) seats in legislatures around the country, the Democratic Party has a weak bench from which to groom future party stars for higher office. Table 5 lays out the comparable losses — and in some cases victories — of other modern presidents compared to the outgoing one.

Table 5: Down-ballot wins and losses for modern presidents

Notes: *Obama’s gubernatorial starting point includes Gov. Janet Napolitano (D-AZ), who was appointed secretary of homeland security in January 2009. Jan Brewer (R) became governor. The 2016 data includes results based on current vote counts for some contests that have not been called. Senate data do not include independents or members of third parties caucusing with president’s party. Having become states in 1959, Alaska and Hawaii’s data are not included prior to John Kennedy. Until Minnesota changed its law in 1973, Minnesota and Nebraska had technically nonpartisan legislatures (Nebraska still has one today). Therefore, Minnesota’s state legislative data are not included prior to Ronald Reagan, and Nebraska’s data are excluded throughout.

[*] Jan. 4, 2017: Correction — originally calculated based on 57 Democratic senators after the 2008 election minus 48 after 2016. But the 2008 figure does not include the two independents who caucused with the Democrats at that time, so neither should the 2016 figure, meaning it should be calculated with 46 Democratic senators after 2016. Thus, 57 minus 46 is 11.

Well, what can we say — we blew it.

We thought the signs pointed to Hillary Clinton winning the White House. We thought that even if she lost Florida, North Carolina, and Ohio, her Midwestern “firewall” of states that not only had voted for Barack Obama twice, but hadn’t voted for a Republican since the 1980s, would hold for her. It didn’t — Trump blew a hole in what we dubbed “Fortress Obama.” Remarkably, this all happened while Clinton was winning Virginia by a larger margin than Obama did in 2012 and almost certainly winning the national popular vote.

Every two years, we put out an update after the election asking, “How did we do?” Well, let’s see:

President

Do we really have to get into it? OK, fine.

We wrongly insisted for months that Clinton was always leading the race and never put her below 270 electoral votes. As of this writing, Trump won 279 electoral votes to Clinton’s 228, according to NBC News projections. We missed the following Leans Democratic states: Florida, North Carolina, and Pennsylvania. We had Wisconsin as Likely Democratic, yet Trump also carried it. Two other Leans Democratic states — Michigan (where Trump leads) and New Hampshire (where Clinton leads) — remain uncalled, as well as Arizona, where Trump leads and we rated as Leans Republican.

Senate

Our big miss was Sen. Ron Johnson’s (R) reelection in Wisconsin (we had him as an underdog for more than a year). Additionally, we did not pick Sens. Kelly Ayotte (R-NH) and Pat Toomey (R-PA) to hold on — Toomey won a close race and Ayotte is narrowly trailing and may be headed to a recount in New Hampshire with Gov. Maggie Hassan (D). We picked a 50-50 Senate; it will be 52-48 or 53-47 Republican. We picked the other 31 races correctly with New Hampshire outstanding.

Governors

Remarkably, Trump rolled in West Virginia and Clinton dominated Vermont, yet both states elected governors from the other party, as predicted here. We missed Indiana’s open seat race, where Lt. Gov. Eric Holcomb (R) will replace Vice President-elect Mike Pence (R), as well as New Hampshire’s open-seat contest, where Executive Councilor Chris Sununu (R) narrowly defeated fellow Executive Councilor Colin Van Ostern (D). Overall, we got nine of 11 right, with one remaining uncalled race in North Carolina — we picked state Attorney General Roy Cooper (D), who is slightly ahead.

House

Given the size of the Republican victory nationally, we overshot on a Democratic gain of 13 seats. Currently Democrats have netted six seats, according to Politico, and several races remain uncalled in California, where votes will trickle in over the next several weeks. We’ll update the final tally in the coming weeks.

We heard for months from many of you, saying that we were underestimating the size of a potential hidden Trump vote and his ability to win. We didn’t believe it, and we were wrong. The Crystal Ball is shattered. We’ll pick up the pieces starting next week as we try to unpack what happened in this election, where there was so much dramatic change from just four years ago.

We have a lot to learn, and we must make sure the Crystal Ball never has another year like this. This team expects more of itself, and we apologize to our readers for our errors.