Thursday, November 10, 2016

Turnout or Vote Margin? How Did Hillary Clinton Lose?

My first post on this blog in years comes on the day after Donald J. Trump won a stunning election over Hillary Clinton and made us all come to grips with the coming horror show that will be a Trump administration. Democrats, college-educated whites, and minorities all over America are wondering, how did we fuck this up? There are plenty of takes on this issue from just about every media outlet around the world today about scandals, political vision, polling, and the media's role in this outcome, but I'm going to focus very simply on one question: what does a Trump electorate look like? Who is responsible for this mess? 

I was listening to NPR this morning, and I heard one of the hosts ask a guest about minority turnout being down. The guest pushed back very hard against this narrative about minority turnout being to blame and focused instead on the white vote margins, which broke hard for Trump.  So I looked into the exit polls from last night to come up with some answers and see which of these perspectives is right. It turns out that with only a couple of exceptions, it was the overwhelming and sudden turn towards conservatism by white voters that governed this election. 

My analysis focuses exclusively on race and the share of turnout in the last four presidential elections in exit polls for 25 states in four different regions of the U.S. 

White turnout

White turnout in the Northeast as a percentage of the overall electorate was mostly down, continuing the demographic trends of past elections. The big exception to this though was Pennsylvania, where the white share of voters increased from 78% in 2012 to 81% in 2016. This was the difference between a 1 point Clinton win and a 1 point Trump win.    

In the South, white turnout was again mostly down as a percentage of the overall electorate. The exceptions were North Carolina and South Carolina, though neither increase by itself was enough to have an effect on the outcome.

The decreasing trend from 2012 to 2016 in white share of the electorate seen in other regions was not seen in several key states in the Midwest. The white share of the vote in Wisconsin, Ohio, and Missouri was not lower than in 2012, and Trump won all three of those states. However, the white share of the electorate was still down from 2004 and 2008, when Democrats won those states. White turnout alone in the Midwest was not the reason that Trump won. 

White turnout in the West has been shrinking more rapidly than in other parts of the country, but the trend slowed down in 2016 or reversed itself altogether in the case of Arizona and Washington. I think it's interesting that over the past 12 years, Arizona's and Nevada's electorates have diverged, so that now Nevada's electorate resembles Las Vegas, while Arizona's electorate now more closely resembles Colorado's. Arizona was always a stretch for the Clinton campaign, and to turn it blue, Arizona's electorate would have had to have been much less white. That didn't happen.

Black Turnout

In New York, the black share of the electorate was stronger than it has ever been, at over 20%. In New Hampshire and Maine, black people don't exist. But in New Jersey and Pennsylvania, the black share of the electorate fell by four and three percentage points, respectively. This, combined with the white turnout mentioned earlier, turned Pennsylvania from a 5 point Obama win to a 1 point Trump win. (It also dropped the Democrats' margin in New Jersey by 5 points, from an 18 point Obama victory to a 13 point Clinton victory.)

But Pennsylvania was only one state, and the mechanics of the Trump win were different in other states.

As a share of the total turnout, black turnout in the south was as strong as ever in Florida, Virginia and Georgia, but fell a few percentage points in the Carolinas. This is notable in North Carolina because there was significantly less early voting in 2016 than in 2012 due to a law passed by the North Carolina state legislature in 2013. Though the law was struck down by the 4th Circuit Court of Appeals, the early voting restrictions remained in place in many counties, including early voting on Sundays, which hurt black turnout. But black turnout was also way down in South Carolina, which doesn't have early voting, but also didn't have early voting in 2008 or 2004.

Black turnout versus white turnout was an important component of the Clinton loss in North Carolina, providing about 3.5 percentage points of her 4 percentage point loss.

Black turnout as a percentage of the electorate in the Midwest was mostly flat from 2012 to 2016, with the exception of Illinois, where turnout rose from 2012. However, in every state with exit polls in the Midwest, turnout was more diverse than in 2004. 

At Vox, Brad Plumer expounded on a tweet from NPR's Domenico Montanaro that compared Clinton's vote totals for Milwaukee and Detroit to Obama's vote totals from 2012, and argued that this was the reason that Clinton lost Wisconsin and Michigan.
In case that tweet is hard to read, Clinton garnered 129,000 fewer votes in heavily Democratic Detroit than Obama did four years ago — and lost the state by around 61,000 total votes. She also got 95,000 fewer votes in heavily Democratic Milwaukee than Obama did — and lost the state by 73,000 total votes.
Well, it is true that Clinton got 95,000 fewer votes from Milwaukee than Obama did, but Trump also got 30,000 fewer votes there than Romney did. Obama won Milwaukee 67% to 32%, and Clinton won Milwaukee 66% to 29%. Detroit is a different story: 73%-26% for Obama, but only 67%-30% for Clinton as of right now, and the vote margin in Wayne County is actually now 95,000 off from the Obama-Romney totals. 

The narrow margin in Michigan means that there are many potential causes for Clinton's (presumed) loss, and so turnout could be one of them. Clinton is currently down by 0.3% in Michigan, but if 16% of the electorate was black voters like in 2012 instead of the 15% that it ended up at, Clinton's position would have improved by about 0.8%. But my point is, black turnout was only 12% in 2008, and Obama won by 17 percentage points. Something else went down in Michigan.

In the west, the black percentage of the electorate is less than 10% everywhere. Though black turnout fell (especially in California), it wasn't a large enough component of the electorate to make a difference to the presidential race.

Hispanic Turnout

I thought the 2016 election would be a breakout moment for the Hispanic electorate, but while the Hispanic share did rise nationally, the rate wasn't as meteoric as I would have hoped given Donald Trump's stance on immigration and the abilities of Indiana judge Gonzalo Curiel. Still, Hispanic turnout share wasn't down, and barring other changes this should have at least locked in 2012's electoral map. But it didn't.

Latinos are still a small percentage of the electorate in the Northeast, but besides New York, the Hispanic share of the electorate rose or, in Pennsylvania's case, remained steady. In order for Latinos to have swung Pennsylvania to the Clinton column, they would have needed to increase their share of the electorate by two percentage points at the expense of the white turnout over 2012's electorate, and that was probably unlikely anyways.  

The Hispanic population in the South is only large enough to make an electoral impact in the states of Florida and Texas, and it turns out that they showed up to the polls in 2016. In part because of strong Hispanic turnout in Texas, Trump's margin appears to be smaller than Romney's was in 2012. However, that was not the case in Florida, won by Obama twice, which went for Trump despite the Hispanic share of the electorate increasing 5 percent since 2008. Therefore, unlike in Pennsylvania, turnout was not the determining factor in Clinton's loss in Florida.

The Hispanic portion of the electorate was not large enough to have much of an effect on the 2012 election in the Midwest (except in Illinois), and the Hispanic portion of the electorate continued to not be large enough to have much of an effect on the 2016 election in the Midwest (except in Illinois). 

Conversely, Hispanic turnout is very important to Democratic victories in the West. But while California and New Mexico showed stronger Hispanic turnout than in 2012 (in California's case, an increase of an astounding 9 percentage points), the Hispanic share of the electorate was flat or decreasing in Arizona, Nevada, Colorado and Washington. In Nevada, Colorado and Washington, this didn't have an effect on the outcome (though it almost did in Nevada). But in Arizona, an unexpected decrease in Hispanic turnout share meant that Clinton's dreams for winning the state never had a chance.

Overall, with few exceptions, the turnout of minorities was not down and turnout of white people was not up, at least not as a share of the overall electorate. One of the exceptions was Pennsylvania, where minority turnout was decisive, and turnout also played a limited role in the Trump win in North Carolina. But Florida and most of the surprise states in the Midwest were won by Trump not because black and Hispanic people didn't get to the polls, but rather because the white voters who showed up were overwhelmingly more Republican than white voters in past elections. 

White vote margins

For these graphs, an upward movement indicates more votes for the Democrat while a downward movement indicates more votes for the Republican. In the Northeast, more white voters chose Trump than chose Romney. Only in Maine did Clinton win white voters, barely, while in New Hampshire, for the first time since 2000, the Democrat lost the white vote. New Hampshire's very small community of minority voters was just large enough to save the state for Clinton, by a whopping 0.2% as of this writing.

But most interesting to me is the fact that Pennsylvania's white voters did not choose Trump much more than they chose Romney in 2012, when Romney lost the state by 5 points. In Pennsylvania, like almost nowhere else in the country, it really was turnout that made the difference and not the increased conservative nature of white folk.  

White Democratic voters in the South have been in the minority for awhile, and all the points above are well below the zero-line for Democrats, but movement towards Trump was only a big feature in one of these states: Florida. Democrats went from losing white voters by 24 percentage points in 2012 to losing white voters by 31 percentage points, and this made all the difference in the outcome, even though white turnout share declined by five points. 

Notice that North Carolina's white voters actually grew more liberal. While turnout share for white people slightly increased, those voters were also slightly more Democratic than in 2012. While Clinton's loss in North Carolina wasn't entirely due to lower black turnout, it also wasn't even partially due to an increasingly conservative white electorate, as in other states.

Georgia was another stretch goal of the Clinton campaign, but it was always going to require either massive black turnout or a liberalization of the white vote akin to what has happened in Virginia and North Carolina.  But Georgia's white vote was as solidly Republican as ever, so it didn't happen. Texas though got closer.

This is, I believe, the most meaningful chart to explain Clinton's loss of the Midwest and therefore the presidency. The Northeast's white voters plunged more conservative, but they started from a high Democratic Party preference. The South's white voters have been staunchly conservative for some time, but they did not grow more conservative with the Trump election. However, the Midwest has seen its white voters grow rapidly conservative since they elected Obama in 2008. 

Look at 2008. White voters approved of the Democrat in Illinois, sure, but also Iowa, Michigan, Minnesota, and Wisconsin. Ohio wasn't so far behind either. Since that election, white voters in the Midwest turned hard against Democrats, in both 2012 and 2016. White Michigan voters went 51% - 47% for Obama in 2008. Clinton just lost them 36% to 57%! A 25-point swing in just eight years!

It is almost the same story in Wisconsin, Ohio, Indiana and Missouri. White voters moved 21 percentage points more Republican in Wisconsin, 19 percentage points in Ohio, 23 percentage points in both Indiana and Missouri. In 2012, Democrats overcame the drop in Michigan, Ohio and Wisconsin by turning out minority voters, which was enough to keep the Midwest blue for a cycle. But the white vote continued to plummet for Democrats, and it was too much for minority voters to overcome. And it wasn't turnout at all. The electorate was marginally more diverse than in 2012, and certainly more diverse than in 2004. It was just that white voters are voting for Republicans by much larger margins than they used to. 

However the white apocalypse for Clinton was not a national phenomenon, it was just a Midwest / Northeast thing. White voters in the South were more or less the same, while white voters in the West actually turned more toward Clinton than they had for Obama in 2012, for the most part. White voters actually picked Clinton in California and Washington, while the margin was close enough in Colorado and New Mexico that Latino voters were able to overcome the margins and give Clinton comfortable wins in those states. Arizona was competitive for so long even though white turnout was up only because white voters moved 21 percentage points closer to the Democrat. 

Nevada almost went to Trump because it was the one western state (with exit polls) that showed more white voters were voting Republican than in years past. Clinton's win in that state, at 2.5%, will be much narrower than Obama's 7 point victory over Romney.

Black vote margins

The margins that black voters gave to Hillary Clinton were down fairly significantly everywhere from the margins that black voters gave to Barack Obama in both 2008 and 2012. (I don't know what the deal is with Illinois either.) The effect was more noticeable in some states than others, but determinative in almost none due to the fact that black voters don't make up a huge part of the electorate in many states. However, the effect moved the polls almost 2% more towards Trump in North Carolina, so that when combined with the lower black turnout it proved to be consequential. If black turnout in 2016 in North Carolina was the same as in 2012, and if black voters voted for Clinton by the same margins as they voted for Obama in 2012, then North Carolina would have gone for Clinton.  

Georgia, with its 30% black electorate, had the potential to be very close for Clinton if black voters voted for Clinton at the same (rather insane) 98% to 2% margin they voted for Obama in 2008, but very close is not the same as "would have won".

Florida's black voters, though smaller proportional to the electorate than other Southern states, had one of the larger movements towards Trump among black voters nationally, and had they voted at the same rate for Clinton as they did for Obama in 2012, Clinton's position would be 1.5 percentage points better. That is just about the margin by which Trump will beat her in Florida.

Hispanic vote margins


Even with all the anti-Mexican rhetoric from Trump on the campaign trail, Latinos voted for Hillary Clinton at lower margins nationally than they had for Barack Obama in 2012. 

While the margins that black voters gave Democrats were pretty much the same everywhere, there was a great diversity to the Latino vote depending on the state. However, the small size of the Latino electorate in most states means that the movements aren't very impactful. In Florida and Texas, Hispanics are generally more conservative than Hispanics in the rest of the nation, but Florida's Hispanics have grown precipitously more liberal in the last 12 years, and that trend continued in 2016, so it had no effect on Trump's victory in that state.  The effect of Texas's relatively conservative Latinos amounted to about 4 percentage points for Trump, but Trump won Texas by 9 or 10 points, so this isn't determinative of the result. 

In the West, Latinos actually moved more in Trump's direction than they did in Romney's in 2012 in Colorado, Arizona, Nevada and New Mexico. Clinton still won all those states but Arizona, which put a particularly sharp decline in the Democrat's margin from 2012. If Hispanics in Arizona voted for Hillary Clinton at the same vote margin that they voted for Barack Obama in 2012, Clinton would have done 3 points better in Arizona. However, she's losing by about 4.5 points now.


CNN looked at the national exit polls to do an autopsy of the Clinton loss, but they placed the blame on "African American, Latino and younger voters". But while it is true that turnout overall was down, Clinton lost primarily because white voters in the Midwest (and Florida) have grown substantially more conservative. States that Obama won in 2008 that Clinton lost included Iowa, Wisconsin, Michigan (probably), Ohio, Indiana and Florida, and in all these states, the biggest electoral effect came not from minority turnout but rather from a more conservative white electorate. Only in two states won by Obama but lost by Clinton (North Carolina and Pennsylvania) did minority turnout have a greater effect on the election than the white turn towards Trump.