How did Donald Trump win the presidential election? In Part 1 of this four-part series, I showed how Trump had performed historically poorly by the standards of challengers in post-incumbent elections (i.e., elections with no incumbent on the ballot following the reelection of an incumbent). In Part 2, I applied the same analysis to show that Trump also didn’t do especially well in the battleground states but won because the handful of states he flipped narrowly from the Democrats were almost all large states. In Part 3, I looked at turnout — specifically, what percentage of the voting-eligible population voted for each candidate — and concluded that, while Trump did better than past GOP nominees in some states, he still underperformed and depended heavily on declining Democratic turnout.
Today: why it’s reasonable to believe that another Republican nominee in 2016 would have collected more of the votes Trump left on the table, and some conclusions about the relative importance of trends, candidates, campaigns, and demographics.
Where Trump Left Republican Votes on the Table: The Senate
We can’t directly compare Trump to another Republican presidential nominee. He’s the only one who won the nomination, and the whole environment of the campaign would have been different with a different nominee. The only head-to-head comparisons we have are from the polls conducted during the primary, in which Trump ran terribly against Hillary Clinton, while Ted Cruz ran even with her and Marco Rubio and John Kasich ran consistently ahead, not just nationally but in the critical battleground states. Those polls are one useful apples-to-apples data point, but they also ended up badly underestimating how many Republican voters who opposed Trump in the primary ended up coming home to support him in November. So while we can conclude that Trump came out of the primaries with higher unfavorable ratings, and he appealed to a smaller potential universe of voters than his main rivals did, we can only make an educated guess about how a different nominee would have performed.
But what we can conclude is that Trump failed to win the votes of quite a lot of people who were at least open to voting for Republicans for important positions. The next best thing we have to a comparison at the presidential level is a comparison of Trump to down-ticket Republican candidates in Senate, House, and governorship races (one of whom, Rubio, had run in the primary). None of these are perfect comparisons, either: Each candidate was running in a single state or district against a different opponent, and many ran with the advantage of incumbency, or against an incumbent. The governors are the weakest parallel, as the ability to separate the race from national dynamics is probably why the Republican candidate won the Vermont governor’s race, while the Democrat won the West Virginia governor’s race.
House Republican candidates as a whole ran ahead of Trump, winning the popular vote by 1.38 million votes, 49.1 percent to 48.0 percent. Republican candidates also ran ahead of Trump in most of the contested Senate races, although California had no Republican in its Senate race, and in the second-largest state with a Senate race, New York, Trump ran 800,000 votes ahead of Wendy Long, Chuck Schumer’s hapless opponent. Let’s look at the Senate comparison, excluding California but including Louisiana’s Election Day “jungle primary” featuring nine Republican and seven Democratic candidates. The first set of columns compares Trump’s popular vote to that of Senate Republican candidates; the second compares them by the two-party vote:
Most of the winning candidates ran ahead of Trump, especially in swing-state races, although Republicans in two contested races (Nevada and Colorado, the latter a race against a Democratic incumbent) ran a bit behind Trump, and in two others (Indiana and Missouri), they ran significantly behind. The fact that Indiana and Missouri GOP Senate candidates won but did so running well behind Trump is best explained as a result of Trump’s popularity in those states and of having Democratic candidates who were better suited to their state’s electorate than Hillary was. Trump ran behind the Republican candidate in popular-vote percentage in 23 states, and he ran behind the Republican candidate in the two-party vote in 21 states. Overall, he ran 1.1 percent behind Republican Senate candidates, 0.2 percent behind in the two-party vote. Outside New York, that widens to a two-point split in the popular vote and a one-point split in the two-party vote.
That looks like a small difference at the margins. But the statewide vote actually conceals the fact that Trump and conventional Republicans running for Senate and House seats had different coalitions that didn’t entirely overlap. When you drill down to the county level, Trump left a lot of potential Republican votes on the table — while winning a significant number of voters who didn’t usually support statewide Republicans.
Using the county-level data for the presidential and Senate races from David Leip’s U.S. Election Atlas, I compared Trump’s vote totals in each county to the vote totals for the Republican Senate candidate in that same county, in order to isolate the counties where Trump got fewer (or more) total votes than the Republican candidate for Senate. We can consider the extra votes for the Senate candidate to be potential Republican voters that a different presidential nominee might have won, or at least contended to win. On the flip side, we can consider the extra votes for Trump to be a sign of his own appeal to voters who weren’t attracted to more traditional Republicans.
I left off Louisiana, since the county-level data that was available was for the runoff, and California, with no Republican Senate candidate. It’s important here to mention that there are almost always more votes cast in the presidential election than in the Senate or House races conducted the same year. While some disgruntled voters might leave the presidential line blank (in Michigan, there were significantly more of these than the margin separating the two candidates), it’s more common for people who followed the presidential race to ignore the rest of the ballot. The “P Overvote” column on this chart illustrates that 1.8 million more votes were cast in the presidential race than in these Senate races. Thus, if Trump simply ran even with down-ticket Republicans on a percentage basis, we’d expect him to be somewhere in the neighborhood of 900,000 votes ahead of them. For example, Trump got 19,031 more votes than Pat Toomey, but there were 111,156 more votes cast on the presidential line in Pennsylvania than in the Senate race. That’s why Toomey’s margin of victory was still bigger than Trump’s.
Instead, if you look at the “net” column, you will see that Trump wasn’t ahead: He was 23,662 votes behind, and more than 800,000 behind outside of New York. Six Republican Senate candidates drew at least 100,000 more votes than Trump, and three (Rob Portman, Marco Rubio, and Mike Lee) drew over 200,000 more.
But that’s just the statewide totals. At the county level, Trump drew more votes than the Republican candidate for Senate in 1,211 counties, and fewer in 765 counties. Trump drew more votes than the Republican candidate for Senate in every county in four states. One was Missouri, the other three were states with Senate candidates with little money or name recognition who got massacred: Oregon, Connecticut, and Hawaii. Trump also got more votes than Wendy Long in 61 of New York’s 62 counties, the sole exception being Trump’s home of Manhattan. On the other end of the scale, Trump got fewer votes than Mike Lee in every county in Utah, and fewer votes than John Hoeven in every county in North Dakota. Trump also got more votes than John Thune in just one of South Dakota’s 66 counties and more votes than Chuck Grassley in just three of Iowa’s 99 counties.
But in states where there was more variation between counties, we see the same pattern over and over: Trump won a larger number of votes in more counties, but that’s outweighed by the Republican candidate for Senate getting a larger number of votes in the larger, more populous urban and suburban counties. All told, across the 765 counties in 28 states where Trump got fewer votes than the Republican candidate for Senate, he received 2.176 million fewer votes.
Trump ended up winning more than his share of the states where the presidential election was relatively close. He won ten of the 17 states decided by less than 10 points, six of the ten states decided by less than 4 points, four of the six states decided by less than 2 points, and three of the four states decided by less than a point. The only really close states he lost were New Hampshire (by 0.37 percent, or 2,736 votes); Minnesota (by 1.52 percent, or 44,765 votes); Nevada (by 2.42 percent, or 27,202 votes); and Maine (by 2.96 percent, or 22,242 votes). But had things gone a little differently, those 2 million votes could have made a decisive difference in a number of states. The column marked “% of State) divides the “SEN R+” column (how many fewer votes Trump received than the Republican Senate candidate in the counties where he got fewer votes) into the state’s presidential electorate. Thus, for example, the 296,541 more votes for Portman than Trump in the 63 counties where Portman got more votes equals 5.36 percent in the Ohio presidential contest. I highlighted in yellow every state where the missing Republican Senate votes cost Trump at least 2 percent of the vote. It’s not hard to picture a future election in which Trump could use an extra 4.7 points in Arizona, or 3 points in Wisconsin, or 2.8 points in Florida, or 1.6 points in Pennsylvania. Already, the margin by which Trump trailed Kelly Ayotte’s vote in six counties in New Hampshire was three times the size of Hillary Clinton’s margin of victory in the state.
The numbers above, however, are before we factor in the overvotes, below:
924,712 more votes were cast on the presidential line in those 765 counties. If we arbitrarily assigned half of the overvotes to Trump (a highly variable assumption by state and county, but this is a back-of-the-envelope measure, so using something more tailored would give us a false impression of precision), the actual net number of Republican Senate votes in these counties by people who didn’t vote for Trump would be closer to 2.6 million.
The flip side of that, of course, is the counties where Trump drew more votes than the Republican Senate candidate. There were 884,545 overvotes in those counties, so using the same “haircut” reduction of half of the overvotes, Trump still drew somewhere in the neighborhood of 1.7 million more votes in these mostly rural counties. For all his weaknesses as a candidate, it’s not hard to locate evidence of a meaningful number of Trump voters who are not otherwise Republican voters.
The irony is that Trump is probably the most urban candidate ever to run for president; he’s lived his whole life in New York City and made his living building apartment buildings, hotels, and casinos, almost exclusively in urban areas. Yet, the places where he ran better than other Republicans were almost entirely in rural and small-town America. Florida is a perfect example: Marco Rubio won 200,000 or more votes in eight counties and drew at least 13,000 more votes than Trump did in seven of the eight, including 86,064 more votes in Miami-Dade County alone. Trump got more votes than Rubio in 44 counties, to Rubio’s 23, but Trump netted 52,966 more votes from those counties compared to Rubio’s extra 270,271 votes from half as many counties. Arizona is even more dramatic: John McCain’s margin over Trump was almost ten times as many extra votes from half as many counties. One rare exception was Clark County, Nev. (Las Vegas), where Trump ran 16,323 votes ahead of Joe Heck. Heck won every other county in Nevada, including Washoe County (Reno, which Trump lost), but his 82,445-vote margin of defeat in Clark County was his undoing in a race he lost by 26,915 votes.
There were 24 counties where Trump received at least 20,000 fewer votes than the Republican Senate candidate:
This list is dominated by cities: Miami, Phoenix, Orlando, Chicago, Seattle, Milwaukee, Atlanta, Salt Lake City, Des Moines, Cincinnati, Cleveland, Columbus, Charleston, Tuscon, Provo, Raleigh, Overland Park. There are also some big suburbs, such as Montgomery County in the suburbs of Philadelphia. More than 13 million votes were cast in these counties, and Trump won 979,082 fewer votes than the Republican Senate candidates below him on the ballot.
Most of these counties, outside of Illinois and Washington, were in states Trump won. Trump carried six of the 24 counties, compared to ten for the Senate candidates. Portman won Hamilton County (Cincinnati) 54–42; Trump lost it 53–24. Grassley won Polk County (Des Moines) 52–44; Trump lost it 52–40. Tim Scott carried Charleston County 56–41; Trump lost it 51–43. But even where the county-level outcomes were the same, the margins were very different. Rubio lost Miami-Dade 54–43; Trump lost it 63–34. McCain won Maricopa County (Phoenix) 55–40; Trump won it 48–45. Johnny Isakson lost Fulton County (Atlanta) 59–37; Trump lost it 68–27.
If we expand our view to counties Trump lost by 10,000 or more votes, we get a much longer list of 52 counties, in which more than 21 million votes were cast and Trump got 1,377,179 fewer votes than Republican Senate candidates:
There are a few more counties here that Trump lost while the Republican Senate candidate was winning, including Cobb County. Georgia in the Atlanta suburbs (Isakson won it 53–42; Trump lost 48–46), Chester and Bucks Counties, Pennsylvania in the Philadelphia suburbs (Toomey won Chester 49–47 and Bucks 52–47; Trump lost Chester 52–43 and Bucks 49–48), and others in Iowa, Florida, and Ohio.
By contrast, few of the counties where Trump got more votes than the Republican Senate candidate were large enough for him to outdistance them by 10,000 votes, let alone 20,000. Trump ran 20,000 or more votes ahead of the Republican Senate candidate in just 13 counties, ten of them in New York State (the others were New Haven and Hartford Counties in Connecticut and Honolulu County in Hawaii; Trump ran 39,274 votes ahead in New Haven). However, his single largest advantage over a Senate candidate was even larger than his deficit in Maricopa: He ran 103,127 votes ahead of Wendy Long in Suffolk County on Long Island, and another 80,376 votes ahead in neighboring Nassau County, as well as 70,000 votes in Erie County (Buffalo), 57,000 in Kings County (Brooklyn), 50,000 in Queens, and 38,000 in Richmond (Staten Island). That is partly explained by Trump’s standing as a hometown candidate, but more powerfully explained by how uncompetitive Long was with Chuck Schumer.
Trump ran 10,000 to 20,000 votes ahead in another 22 counties, nine of them in New York and six others in Oregon and Connecticut. The rest were four in Missouri plus Luzerne County in Pennsylvania (12,137 votes, Trump won 58–39 compared with 51–43 for Toomey); Madison County, Ill. (13,825 votes, Trump won 55–39, Kirk lost 50–44); and Clark County, Nev.
Where Trump Left Republican Votes on the Table: The House
Sixteen states didn’t have Senate races, including some hotly contested presidential battleground states (Michigan, Minnesota, Maine, Virginia) and the two states where Trump won his largest total popular-vote margins (Texas and Tennessee). Then there’s California, where there was no Republican Senate candidate. To supplement the lessons from the Senate races, let’s consider how Trump’s vote haul compared to the House races in those states. I looked at all the presidential battlegrounds and every state large enough to cast at least 2.5 million votes in the presidential contest, a total of 30 states. The results:
House races are even harder to compare to the presidential race because we don’t yet have comprehensive data at the district level, and there were 47 uncontested races in these states (31 unopposed Democrats and 16 unopposed Republicans). But overall, there were almost 6.4 million more votes cast in the presidential race than in House races across these states, yet Trump received 204,495 fewer votes than the Republican House candidates. That number quadruples outside of Massachusetts, where Democrats won all nine House seats, five of the nine unopposed. In 18 of these 30 states, Trump received fewer votes than the Republican House candidates; in 22 of 30, he got a lower percentage of the vote. Trump got 1,394,441 fewer votes than Republican House candidates across just eight states: California, Texas, Florida, Illinois, Pennsylvania, Ohio, Georgia, and Washington.
In some states, the absence of Republican candidates from the ballot makes the disparity even more dramatic. Trump got 73,567 fewer votes than House Republican candidates in Virginia even though there were 201,183 more votes cast in the presidential race in Virginia than in the House races, and even though there was no Republican candidate in Virginia’s eleventh district. There were 282,003 votes cast in the VA-11 House race.
In California, Trump got 198,223 fewer votes than House Republicans even though there were nine districts with no House Republican candidate (in which 1.7 million votes were cast) and 767,577 more votes cast in the POTUS race than in House races.
The picture from the battleground states (the ones decided by fewer than 10 points) is mixed. Trump ran significantly behind House Republicans in Georgia, Ohio, Pennsylvania, Texas, Florida, Colorado, Virginia, New Mexico, North Carolina, Arizona, Maine, and Iowa, behind in Minnesota, and behind in popular-vote percentage but barely ahead in vote totals in Nevada and Michigan. He did, however, run ahead of House Republicans in Wisconsin (where there was no Republican candidate in two of the state’s eight districts) and New Hampshire.
Where Trump Left Republican Votes on the Table: The Hispanic Vote
The demographics of voters who pulled the Republican lever in 2016 but did not vote for Trump is a broader question; the most obvious example is the Mormons, a reliable Republican-base voter group that abandoned Trump in droves. But it’s impossible to pass this topic without noticing that one of the specific ways in which Trump seems to have run behind other Republicans was with Hispanic voters.
There’s a huge methodological debate over how to count Hispanic voters and whether exit polls in particular captured them accurately. For example, the widely cited Latino Decisions polls, which had Trump doing a lot worse with Hispanic voters than the exit polls showed, are conducted in both Spanish and English, and they select Hispanic voters by looking for Hispanic surnames rather than self-identification. But even if the exit polls aren’t necessarily the last word, they offer an apples-to-apples comparison of Trump to down-ticket Republicans among the same sample of self-identified Hispanic voters. What they show is that Trump ran well behind several Republican candidates for the Senate or the governorships:
All things considered, 2016 was not the extinction-level event for Republicans among Hispanic voters that many people had expected. But while there were variations from state to state, it seems apparent that Hispanic voters were willing to draw distinctions between Trump and other Republican candidates. None of the candidates for whom we have exit-poll data actually won the Hispanic vote, but Rubio, Richard Burr, and Utah governor Gary Herbert were highly competitive. Others, such as McCain, Isakson, and Mike Lee also did respectably by running double digits ahead of Trump. In Burr’s case, he also ran double digits ahead of Pat McCrory. If you’re looking for voters who were open to a Republican message and a Republican vote in 2016 but didn’t vote for Trump, this is one group to examine more closely.
Conclusion: What We Learned
Many election analysts and pundits predicted that Hillary Clinton would swamp her Republican opponent, especially once it proved to be Donald Trump, because of the demographic trends that had provided Barack Obama with his margin of victory over Mitt Romney in 2012. Some of us saw things differently and predicted that the historical post-incumbent trend would make Republicans competitive or even the favorites in 2016. Of all the predictions I made in the course of this election cycle, the very first — my original September 2014 study of post-incumbent elections — stands up the best.
But like many people who were hopeful about Republican prospects entering the primaries, I was also emphatic from the beginning of the race until almost the end that Donald Trump could not possibly win it. I put my faith in the importance of candidates, and Trump’s historic and well-earned unfavorables made him look for all the world like the kind of candidates who routinely get destroyed in winnable statewide races. Even at the very end, when the polls showed enough tightening that I was warning that a Trump victory was possible, I still didn’t think it would actually happen, largely because I believed that campaigns mattered — that Trump’s lack of a get-out-the-vote operation and the Democrats’ presumed and widely reported superiority in the ground game would make the difference in a close race. This, too, proved wrong, perhaps because outsiders couldn’t see the scope of the RNC’s stand-in for the campaign ground game; outsiders also failed to see the hubristic ineptitude of the Hillary camp, especially its neglect of the Midwest in pursuit of a national wave and elusive pickup opportunities in Georgia, Arizona, and Texas.
In the end, the historic trend away from the incumbent party was the most powerful factor of all. That doesn’t mean that demographics, candidates, and campaigns didn’t matter: When placed in proper context, we can see that Trump ran well behind the historic averages and down-ticket Republicans, losing more votes from the Republican tent than he brought in.
Would another Republican, or another Democrat, have done better? Trump won every single state carried by Ted Cruz or John Kasich except for Maine, where Trump carried one of the two House districts and greatly improved on past Republican performance. The only state he lost in both the primary and the general election was Minnesota. Hillary Clinton, by contrast, lost twelve of the Bernie Sanders states (Bernie also won Maine), including crucial swing states Wisconsin and Michigan, and she probably lost the popular vote to Bernie in Iowa before losing Iowa in November.
Hillary was clearly a bad fit for hard-to-poll white working-class voters in Michigan, where she was surprised by Bernie after leading him in the polls, and was surprised again by Trump after leading him in the polls. That may or may not have mattered as much against a Republican who didn’t sound like an economically populist Democrat. The declining turnout for both parties in November in Wisconsin may have been a sign that both nominees were bad fits for the state, where each lost the primary by 13 points.
The primaries were also an early warning of Trump’s weakness in urban and educated suburban areas: He lost Manhattan to Kasich; Kansas City, Milwaukee, and Raleigh-Durham to Cruz; and Atlanta, Charleston, Minneapolis, Little Rock, Miami, and the D.C. area to Rubio. They were also a warning that Trump would probably not run as strongly in the western half of the country compared with the eastern half, although the contested portion of the primaries ended before Trump got to the three West Coast primaries.
I suspect that Trump’s unconventional campaign was the only one that could have swung Michigan for the Republican ticket, but I can’t prove this conclusively. He ran only slightly behind House Republicans there.
But there are signs that another Republican could have won the other states that decided the election, quite possibly by a wider margin, and maybe could have won other states such as Minnesota, Colorado, Virginia, or New Hampshire. Iowa, Ohio, and Wisconsin all swung hard to the Republican side in 2010, 2014, and 2016 across the board. In Ohio, that reflected a resurgence of the state’s Republican tradition and recovery from scandals of the 2005–2008 years, as well as the collapse of the Ohio Democrats. Kasich has built a powerful Republican machine there. In the Upper Midwest, demographic trends have lent a hand: In 2004, Iowa, Wisconsin, and Minnesota were among the few states in which the oldest white voters were the most liberal, and the generation born of the Great Depression has been dying off. Trump won none of those states in the primaries (Wisconsin and Minnesota were both double-digit losses, to Cruz and Rubio), and Rubio polled particularly well head-to-head in both Wisconsin and Minnesota. And, unsurprisingly, Rubio ran well ahead of Trump in November in Florida.
If Trump wants a second term, he’ll need to figure out how to win over urban, suburban, and Hispanic voters who voted for other Republicans in 2016.
Pennsylvania is another story: Trump really did catch fire in the rural parts of the state, both in the primaries and in November, running far ahead of Toomey, and he outdistanced Bush, Romney, and McCain in the state. But a Republican less repellent to upscale suburbanites might have matched Toomey’s much stronger showing in the Philadelphia suburbs. Unlike Arlen Specter, who ruled that region, Toomey showed that a conservative, pro-life Republican could win there in the right year. Rubio, Kasich, and possibly Cruz could have done the same.
On the other hand, it seems highly likely that any of the other Republican candidates would have run stronger in Texas, Georgia, and Arizona. But Trump’s bleeding in those states didn’t end up affecting the outcome.
We will never know for sure. What we do know is that Donald Trump had the wind at his back and did some things Republicans had not done for a long time. In the final analysis, though, he did less than the historical trends suggested were possible. If he wants a second term, he’ll need to figure out how to win over urban, suburban, and Hispanic voters who voted for other Republicans in 2016. If he wants to hand the White House to a successor, he’ll need to expand his coalition enormously.