Where Predictions went wrong
I doubt it has escaped the attention of any of my readers that my predictions for Tuesday were off, both in the Presidential race, where Donald Trump won a 306-232 victory and in the Senate, where Republicans held their losses to two seats. It is of course incumbent upon me to ask the question of why? After all, you are not reading this site to get regurgitated conventional wisdom. Despite the redesign, there are much flashy places to get that, and here you miss out on the tone of sneering condescension with which it is delivered by Vox. So what went wrong?
The first suspect, in fact the usual suspect, not to mention the one easiest to blame, is the polling industry. Real Clear Politics' final national average had Hillary Clinton ahead of Donald Trump by 3.2% points. Obviously that was misleading. But not for the reasons everyone suggests, namely that Clinton's lead pointed to her winning. Far more important to me was the direction of that average, which in the final few days moved from a near tie to a substantial Clinton lead. This was evident in the Washington Post/ABC news tracking poll which went from a 1% Trump lead a week before the election to a five point Clinton lead on election day. This movement was echoed by nearly all pollsters regardless of their absolute margins, and that persuaded me that even if they were off, the movement was in Clinton's election, and she should overperform on election day. That did not happen.
This is a much more serious failing for pollsters than merely getting the result wrong. The actual national popular vote margin is likely to end up around Clinton +1.5%, which means that polls showing Clinton ahead by 4 were closer than the single one showing Trump ahead by 3. But historically, even if pollsters are wrong about the relative standing of candidates, they have been able to measure the trajectory of changes in support. The fact that it appears Trump won late deciding voters across the country indicates not merely that the national polls were off, which they were by less than in 2012, but rather that they showed movement that did not exist.
The place where polling actually failed was at the state level. But here again the picture is more complicated than it might seem.. Yes polling was off substantially in North Carolina and Ohio(in favor of Hillary Clinton), and Nevada and Colorado(in favor of Trump), but polls showed a close race in Florida and that is what happened. Where polling really failed was in Michigan, Pennsylvania, and Wisconsin, and the problem there was not that polling got it wrong, some pollsters did very well, but rather that so few bothered to poll those states that the averages never converged on a more accurate picture. Combined with skepticism about Trump's chances, this is the fifth election in a row in which Republicans have claimed to be on track to win Pennsylvania and Michigan in the final days of a Presidential campaign only to come up short, and the incomprehensibility of a two time Obama voter supporting Donald Trump, what evidence there was happened to be dismissed by the Clinton campaign, the media, and too heavily, by myself.
The state where I made the greatest error in terms of outcome was Florida. Prior to the election I indicated that I would be looking at two things. First, Hillary Clinton's performence in Hillsborough, which in recent years has served as a proxy for who will win Colorado and Virginia. Second, the turnout in South East and Central Florida. In the end, both went in directions that were favorable to Clinton, and both actually "worked" as indicators as far as that went. Clinton won Hillsborough 52-44, by a greater margin than Obama did in 2012(52-46), and she did then go on to perform very well with similar voters in Colorado and Virginia. She loos to win the former in the end by more than 4%, and the latter by a bit over 5%. The problem was that while in almost any previous election, winning those states would have been for a Democrat to win the Presidency, this year they did not matter, because while winning those two states that went for Bush in 2004, along with Nevada which also voted for Bush but which I suggested(correctly) was locked down due to early voting, Clinton managed to lose three states which voted for John Kerry that year. More on that in a moment, but to finish up on Florida, there two the traditional indicators failed to function. Clinton's loss in Florida cannot be blamed on low turnout. Clinton won 250,000 more votes in the state than Obama did in 2012. She swept Miami-Dade County bu a historic margin. But Donald Trump, even in the face of Gary Johnson, won 445,000 more votes than Mitt Romney.
Where did those votes come from? For all the talk about the missing white voter that Trump was going to bring to the polls, the numbers showed little evidence that they actually existed. Turnout ended up about 900,000 votes, or 11%, slightly below my expectations based on the 36% increase in early votes cast, but virtually of that increase was concentrated in heavily hispanic or democratic areas. Not just Miami Dade, but heavily Puerto Rican Orange County set records both for turnout and democratic margins, going for Clinton by 26% when Bush had won it as recently as 2004. So, if as I, and most people expected, the increase in turnout came from voters registering to support Clinton or to oppose Trump, how did Clinton lose and where did Trump's new voters come from?
Well the answer to that is that they came from Obama voters. While higher turnout in the Democratic strongholds made them go "blue" by historic margins, everywhere else Democrats collapsed. This was true on the Atlantic Coast where swingy counties went for Trump by more than twenty, but also in the north, where the question as to whether it was possible for Clinton to do worse than Obama's 27-29% was answered when in a number of cases she dropped below 20%. It appears vastly more voters switched to Trump than registered to support Clinton. And in the end that doomed her as illustrated by the maps above(which are taken from David Leip's Atlas of Us Elections, which uses Red for Democrats and Blue for Republicans).
The "Kerry Assumption"
Florida is a polarized place, and was always expected to be close. But the Midwest states delivered a genuine surprise. Ultimately, however, it came from the same thing. Traditionally Democratic voters, who had no issue casting ballots for a Massachusetts liberal and a black guy with a funny name, deciding to vote for Donald Trump. Take a look at these maps of Wisconsin, a state that has not gone Republican since 1988. In 2012, despite the presence of favorite son Paul Ryan on the Republican ticket, Obama comfortably won it, and it sent a Lesbian Democrat to the Senate comfortably over a former Republican governor. While it elected a Republican Senator, Ron Johnson in 2010, his narrow victory over Democrat Russ Feingold was seen as a consequence of laziness on the part of the latter. Feingold's victory in the rematch this year was seen as such a forgone conclusion that both parties pulled out in the spring, and all but wrote it off. At the Presidential level, Clinton did not campaign in the state once after the primary.
Signs that something might be wrong began to surface three weeks ago when the Democratic Senatorial Campaign Committee, having cancelled its ad reservations in the early summer, reentered the race late. This coincided with a number of polls showing the race, which had previously featured Feingold leads of ten points or more, closing to a near tie. Yet still few of those polls picked up the trouble for Clinton. She still lead the polling average for the state by 6 on election day.
It was thus a shock when exit polls came in showing a tie. The state had been close in 2004, Kerry had won it by only 7,000 votes, but that had been against an incumbent President and in the midst of a 3% national sweep. Having already seen the lopsided margins in Indiana, where Clinton had lost by 20, 57-37 compared to, 52-44, for Obama in 2012, I was already pessimistic on her chances. And when Wisconsin came in, I realized for the first time that Trump was likely to be the next President. Because I had seen this map before. In fact I had seen it several times. Scott Walker, the Republican who won the 2010 election for governor has been a figure of hate for Democrats due to his anti-public sector union policies. Combined with Wisconsin's "open" electoral laws, the result has been a series of closely fought battles that resembled the Somme politically. After winning in 2010 by 51-47, a 2011 supreme court race ended up as the first proxy battle, going to the Republican by 0.2%. Then in 2012, Walker faced a recall election, winning 52-45, before winning reelection in 2014, 52-47. In all of those elections, lopsided Democratic margins in the two big counties of Dane(the city of Madison) and Milwaukee, were countered by a collapse in Democratic support outside of them. The cities could cast enough votes to make things close, especially when Dane went 73-24, but never close enough. Obama had managed to halt this trend in 2012, but in 2016, Clinton won the anti-Walker coalition of 2011, 2012, and 2014, rather than the Obama/Kerry coalitions of the last several elections.
In fact, one cannot help but note that the 2016 Presidential map is almost identical to the 2014 Gubernatorial one. It makes no sense then to attribute these defections to events like FBI Director James Comey's letter about Hillary Clinton's emails. This realignment began in 2014. The only difference is that unlike in 2012, the Democrats did not bounce back in 2012.
This then should challenge one of the key myths on which Democrats, or at least the Clinton campaign, based their strategy around, That belief was two-fold. First, that Democrats were purely on the offensive due to demographic change. The gradual decrease in the white proportion of the electorate meant that the real question in American politics was how long it would be until previously Republican states like Georgia or Texas turned Democratic. Central to this was the idea that Democrats had hit rock bottom in 2004. In that year they had run a wealthy Massachusetts liberal against a wartime president fortified by the advice of campaign genius Karl Rove, and in their narrow defeat that year, representing a brief bump on the road of the five out of six elections in a row which they had "won"(counting 2000). As such, it became an article of faith that if John Kerry even managed to win it, then there was no way any other democrat, who presumably had to be better than Kerry, could possible lose it.
In 2006 and 2008 this view seemed to be vindicated. In the former year, Democrats did not lose a single House or Senate seat, while by the end of 2008, no Republican congressman or Senator remained in an Obama seat. Then things began to get ambiguous. In January 2010, Scott Brown won an upset to take a Senate seat in a special election in deeply Democratic Massachusetts, while in 2010, Republicans made massive gains at the state level. But Democrats rationalized these losses as either outliers(in the case of Massachusetts or the Midwest) or as a new bottom. The latter was the explanation relegated to the complete destruction of the Southern and Midwestern Democratic party. In 2010, Democrats lost legislatures in Arkansas, Alabama, Mississippi, Tennessee, Oklahoma, Louisiana, and Mississippi. Not to mention Michigan, Wisconsin, Iowa, and Pennsylvania. Democrats blamed the losses in the former states on race, believing voters had cast ballots against the first black President, and that the latter were the result of temporary discontent.
2012 seemed to provide support for this theory. Obama not only won Iowa, Michigan(against locally born Mitt Romney), Ohio, and Wisconsin, but Democrats won Senate races in Florida, Michigan, Missouri, Iowa, and Wisconsin. The idea that the southern states where Democrats did not recover were somehow "different" and that they would bounce back when Democrats again ran a white candidate became an article of faith. This was maintained through 2014 when Democrats again suffered an even greater wipeout in the Midwest.
In 2016, the main disconnect between the Clinton campaign and rural America was not cultural or gender based, but rather the fact that it was intellectually inconceivable that anyone who voted for Obama twice, and for Kerry could possibly vote for Donald Trump who was "self-evidently" racist, sexist, unfit to be President, et cetera. This, combined with a conviction that Clinton could not possibly do worse in the Appalachian areas where Obama had collapsed into the 30s, because Obama's poor performence had been due to "racism" and Clinton by virtue of being white would win them back, blinded Democrats to what was happening. In the end Clinton did manage to do worse than Obama, not least in her home state of Arkansas. Kerry got 44%, Obama 40% in 2008 and 37% in 2012. Clinton, who had been first lady of the state for a decade, received 33%. In West Virginia, which she won with 70% in the 2008 primary, she received a mere 26.5%. The Clinton campaign never counted on winning either state, but they accepted as an article of faith that they would improve in politically and demographically similar parts of states like Michigan, Pennsylvania, North Carolina, and Wisconsin. In the end, that did not happen. And in hindsight, it is hard to claim there was any reason to believe it should have.
Democrats, however, still have not taken the blinders off. Having convinced themselves that Obama's poor performances were due to racism, they now speculate that Clinton's even worse results were due to sexism.The danger of this is not ,the contempt it conveys, but the fact that it acts as an excuse, by telling Democrats that there is nothing they did wrong, nor is there anything they need to fix. Somehow, these problems will fix themselves and the sorts of people in these states will begin voting for democrats again.
In a later piece I may lay out my thoughts on why Democrats did so poorly in these areas, but for right now suffice to say that these assumptions spread from Democratic campaigns, to the media and even Republican ones, which dismissed Trump's targeting of the Midwest, and failed to recruit strong candidates. I erred on the side of buying those assumptions and will keep that in mind in the future. The question is whether the media or Democratic party will learn anything from it themselves.