In the last article, we looked at how the key to winning a US presidential election is correctly navigating your way around the electoral college, and we ended with looking at the fact that three states (Pennsylvania, Wisconsin and Michigan) swung to Trump after decades of being solidly Democrat. In this article, we’ll look at what happened to cause this.
Our first issue is whether it was more that Trump won the election, or whether Hillary lost it. As Obama won in 2012, Hillary didn’t really need to do anything new, she didn’t need to gain a single state more than he had done and that would make for a comfortable win. As it happened, Hillary lost Wisconsin, Michigan, Ohio, Pennsylvania, Iowa and Florida, without winning any new states. The automatic assumption might be compared to the last time, the total votes for Democrats went down and for Republicans went up. In reality, although Trump ’16 got roughly 2 million votes more than Romney ’12 did, the number of votes for Obama ’12 and Hillary ’16 were almost identical: 65,915,795 to 65,853,514, i.e. Hillary got 99.9% the number of votes Obama four years earlier.
Additionally, Hillary comfortably won in terms of the popular vote, beating Trump by roughly 3 million. So if Hillary won the popular vote, and got the same number of votes as Obama, then the answer lies in the state-by-state changes, and here we see a huge range across the country: at the high end, compared to the number of people voting for Obama ’12, Hillary increased her vote by 23.4% in Utah; on the opposite side, Hillary decreased the vote by 24.9% in North Dakota. Crucially, these swings were not random, so to see how it played out across the country, we can look at the map below.
Firstly, we should not ignore that the good points for Hillary, and a future piece will examine the changing circumstances in places like Texas. However, as this piece is focusing on why Hillary lost, we will only look at where the vote went down. As we can see, 17 of the 50 states saw a substantial drop of at least 10% compared to the previous election, and I’d like to suggest that we can split these 17 states into 3 groups.
1. Unique circumstances
In the first group, we place Hawaii, Vermont and Maine. In Hawaii, one possible reason for the decrease is that Barack Obama was born in Hawaii, and so there was more enthusiasm for a native of their state. When we compare Hillary ’16 to Kerry ’04, Hillary is still up around 35,000 votes (~230k to ~265k), so it’s not so bad. In Vermont and Maine, unlike 2012, which was a 2-way race in each state, 2016 was de facto a 3-way in Vermont and de jure a 3-way in Maine. In Vermont, Hillary got 20,000 fewer votes than Obama ’12, but there were 18,000 write-in votes for local Bernie Sanders. In Maine, Gary Johnson stood for the Libertarian party, receiving 38,000 votes, compared to the 41,000 fewer Hillary got compared to Obama ’12.
All three of these states went blue in both 2012 and 2016; these are interesting changes, but don’t really contribute to the story of why Hillary lost.
2. Deep red states
To start, let’s compare the map above with the map below:
Hillary largely lost votes across the northern states between the coasts, but they show an interesting divide compared to how they voted in 2012. Our second block are those states that are strongly Republican, and consist the northern rockies area (ND, SD, WY, ID, MT), to which we’ll also add Missouri, Mississippi and Western Virginia. These 8 states were primarily strong Republican voters in 2012, and all of them voted Republican in the 5 elections from 2000 to 2016.
There are more nuanced reasons to why each of these saw a fall in democratic votes, but to keep it simple, we can make a safe assumption that a candidate like Hillary Clinton, who represented the epitome of the DNC establishment, isn’t going to be very popular in areas that never vote for Democrats. Either way, all 8 of these states stayed red between 2012 and 2016, so like our 3 ‘unique circumstance’ states, they didn’t cause Hillary to lose.
3. The Rust Belt
For those unfamiliar with the term, the Rust Belt (RB) is a term for an area of the midwest that had high levels of industrial and manufacturing jobs in the post-war boom, but has had a sharp decline in those jobs in the last few decades.
There’s not a strict definition of what does and does not count as the RB, so for simplicity we’ll say it’s the 5 main states that are red in the map (PA, OH, IN, IL, MI) and we’ll stretch it to include Iowa, Wisconsin and Minnesota too.
What’s interesting is when we compare the 2012 vote map with the 2016 democratic change map. Firstly, we can see that this is a region that overall voted comfortably for the democrats in 2012, with seven of the eight (no Indiana) going blue. Secondly, we can see that in another seven of our eight RB states (no Illinois) had a drop in votes, and six (no Pennsylvania) had at least a 10% drop between ’12 and ’16. Thirdly, of the six states that flipped from blue to red and won Trump the election, five are in this region. Finally, even in the two states that did not flip (blue Minnesota and red Indiana), we still saw the same direction of travel as their neighbours, i.e. a Democratic vote drop of over 10%.
So, could it be a coincidence that almost all the states that changed the outcome of the election are contiguous and had the same voting shift? Surely not. A one-layer-deep answer to the answer in the question of this article is that Trump won because the RB went from blue to red. In my opinion, any answer to that question that does not primarily focus on this region simply isn’t paying attention to the numbers.
The above answer, however, still doesn’t address the issue posed earlier: did Trump win because he won the RB, or because Hillary lost it?
To take a slight shift of gears, let’s look at little at the polling for the 2016. Was this loss in the RB something that pollsters saw coming? Simply, the answer is no. Let’s focus on five of the 8 RB states: Indiana, Michigan, Wisconsin, Ohio and Pennsylvania, all of which Obama won in ’08. It depends how you average the polling, but if you go by the aggregated polling data from the popular site RealClearPolitics, the average poll lead for Hillary by election time was 3.6% in Michigan, but Trump won it by 0.3%. They said Hillary led by 6.5% in Wisconsin, but Trump won by 0.7%. They said Trump would win by 10.7% in Indiana but he really won by 19%. Similar errors were made in PA and OH. And that’s just according to RCP – the team at FiveThirtyEight had slightly different polling averages, so, in a number of them Trump’s under-estimation was even bigger.
Now, a simplistic answer to this is ‘oh, well Hillary wasn’t that popular a candidate, so the polls overestimated her popularity.’ But this wasn’t true nation wide. The predictions pollsters made in New York and Texas were pretty much on the money. And the predictions for California actually underestimated Clinton by about 6 points. So we can’t say that this was just a general failure on the pollsters’ part. The same thing as happened with an unusual voting pattern in the RB also happened with the polls there.
So, to understand whether Trump won this area or Hillary lost it, let’s take an in-depth look at our five focus states from above (MI, PA, OH, WI, IN). We’ll start with PA, as it’s overall the outlier.
| 2012 | 2016 | diff | ||
| Pennsylvania | D | 2,990,274 | 2,926,441 | -63,833 |
| R | 2,680,434 | 2,970,733 | 290,299 |
Here we can see that this was more of a Trump win than a Hillary loss – he really did a decent job of increasing his vote share compared to Romney ’12. However, in many ways, this state is split and is only a half-RB state. The reason for this can be seen in the rust belt map above. For one, there is the split between the RB and coal belt. But perhaps more importantly, a large part of the population is in right hand edge of the state. Pennsylvania is home to ~12m people, around half of whom live in the greater Philadelphia area alone. Geographically, Philly isn’t RB – it’s basically part of the East Coast area where Hillary kept or increased her vote.
When we break it down on a county-by-county level this seems to hold. In Philadelphia county, Hillary got 99% the vote that Obama did in ’12. However, the picture is different when we look at the centre and left thirds of the state, which are much more in line with the history and geography of the rest of the RB. If we take five counties at random from across those two thirds of PA (Greene, Beaver, McKean, Huntingdon and Lycoming counties) and examine their Democrat vote change from ’12 to ’16, we see exactly what we saw in the rest of the RB: falls from 12% to 24%, averaging 18% overall. This is a very different picture to the average fall of just 2.1% in Pennsylvania as a whole.
Either way, we can say that in total, Trump won Pennsylvania (but Hillary also kind of lost the RB belt part.)
Next, Indiana.
| 2012 | 2016 | diff | ||
| Indiana | D | 1,152,887 | 1,033,126 | -119,761 |
| R | 1,420,543 | 1,557,286 | 136,743 |
This is a little different: the number that Trump increased his vote by was a little higher, but basically around the same amount as the Democratic vote fell by. The number doesn’t tell us if these people switched, or whether Dems didn’t vote this time and non-voters from 2012 did vote for Trump, but we’ll basically call this a draw: Trump won and Hillary lost.
However, this was a red state that stayed red. Let’s now look at WI, MI and OH. Importantly, these three RB states went from blue to red, and between them carried 44 electoral votes. If they had stayed blue, Hillary would have won 271 to 260 (8 electoral votes went to neither candidate).
First, Ohio and Michigan.
| 2012 | 2016 | diff | ||
| Michigan | D | 2,564,569 | 2,268,839 | -295,730 |
| R | 2,115,256 | 2,279,543 | 164,287 | |
| Ohio | D | 2827709 | 2,394,164 | -433,545 |
| R | 2661437 | 2,841,005 | 179,568 |
In both of these states, the number of people who stopped voting Democrat was in the region of double the number of new Republican votes. In Michigan, if Hillary had only lost 95% of the number she did, she would have won. I think it’s fair to say that these two states were Hillary losses more than Trump wins.
Finally, Wisconsin.
| 2012 | 2016 | diff | ||
| Wisconsin | D | 1,620,985 | 1,382,536 | -238,449 |
| R | 1,407,966 | 1,405,284 | -2,682 |
It’s hard to argue with numbers. Trump lost votes, so it wasn’t that he won the state over at all; Hillary lost over 200,000 voters – not people who changed their mind about which party to vote for, but people who came out to vote for Obama ’12, and just stayed at home when it came to Clinton ’16. This is a clean Hillary loss.
So across the RB, it’s a not a perfectly clean narrative, but overall, it was much more strongly Hillary’s loss than Trump’s win that changed the outcome of the election.
And so we have our two-layer-deep answer to the question of why Trump won: the RB flipped, and it was Hillary’s loss. But it still doesn’t get to the question of why these voters who did vote in 2012 failed to do so in 2016.
At this point, let’s look at some of the reasons that people attribute to Hillary’s losses in the RB region. A bit of googling comes up with a couple of common arguments.
Disclaimer: from here on, both the arguments from others and from myself start straying away from pure data and start mixing in anecdotal evidence and conjecture. I believe everything above is factual, but from now it is more on the side of opinion.
The first form of the argument can be found in this article from Politico – it basically revolves around the idea that Hillary had a terrible ground game in a number of states. It tells stories like a volunteer turning up, asking for a yard sign, being told they don’t statistically change the outcome, and her leaving not to return. Voter data collection was poor, and so they couldn’t react to fix issues because they didn’t know what they were.
There is some merit to the idea that ground game played a factor, but it can’t have been the whole picture. Two reasons against this (and a third a bit later): firstly, I don’t think it rings true on a human level. The 2nd Obama election was a less heated affair: the enthusiasm for Obama was a little down from 2008, and he was running against a pretty moderate Republican. Compare that to the divisive 2016 choice. The idea that people who turned out in 2012 decided against turning out to vote against Trump based on not enough yard signs or enough people knocking on their doors doesn’t seem very plausible, even less so when it would have to attribute for more than 1 in 10 Obama ’12 voters. This can’t possibly be the whole picture. In fact, even in the article, the author hints to the fact this can’t be it. They reference the problem that a lack of data collecting “might have… showed the campaign that some of the white male union members they had expected to be likely Clinton voters actually veering toward Trump”. The underlying problem isn’t the campaign didn’t know that these people were veering toward Trump: it’s that the people were changing their mind in the first place. This argument simply fails entirely to address the reasons why people were changing their minds or becoming disillusioned with Clinton herself.
Argument two is about how the pollsters got it all wrong. The Washington Post complains about a large time gap from 2008 made for poor models, and incorrectly predicting the black turnout. A similar argument from American Progress highlights how they got it wrong in the turnout of non-college educated whites. However, this argument is just a higher-level repeat of the first argument: okay, so what if the polls had better modelling and did get both of those groups correct? They would have called the shift correctly, but that doesn’t explain why the shift happened in the first place.
The final argument is similar to the first, but rather than focusing on the Democratic ground game, it focuses on Hillary herself. She had a limited amount of time to visit places, and didn’t come to states like Michigan enough. Like above, we can question the idea that 10% of Obama ’12 voters failed to vote for her because she wasn’t at some point within a few hundred miles of them. But more importantly, it doesn’t hold up to the data: if you look at this map from the Boston Globe, you can see that actually she spent a huge number of her visits in Iowa, and a decent number in Ohio. Yet the fall in vote was consistent across the region, not correlated to her physical presence. Again, I’m not arguing that in-state visits don’t matter, but to chalk the entire thing up to this neither seems particularly convincing nor is backed up by the data.
So if the most common arguments are not correct, what was the underlying reason that caused people to either switch to Trump, or at least to lose enthusiasm with Clinton to the degree that they did not bother turning out to vote? I would like to suggest that it was about the economy, and more specifically how it ties in with issues around international trade and NAFTA.
Before going into the numbers that I think back my argument up, we’ll first look at little at what NAFTA is, and the history of how trade has affected the RB, or at least how people there perceive it to have affected them.
So what is NAFTA? The North American Free Trade Agreement was the culmination of decades of smaller deals and progress towards a free-trade zone between the USA, Canada and Mexico, and was brought into law in 1994 under Bill Clinton. The effects of the agreement are highly disputed, some saying it was great for America, others that it was terrible, but one indisputable effect was that it led to a lot of jobs being lost in American manufacturing. It’s possible that more jobs were gained that made up for it, but that’s hard to prove either way, and isn’t entirely relevant right now. Let’s focus on the popular idea that factories were closed in the US, and set up in cheaper places like Mexico.
A lot of evidence suggests that perhaps automation played an equal or larger part in the reduction of the role of the manufacturing jobs, but I believe in the public perception, this hasn’t really had an effect on the narrative. Let’s say your town has a 3000 person factory that gets shut down. The owners upgrade and invest in machinery, so when they open the new factory in Mexico, it only employs 1500 people. I would suggest that people are far more likely to say that “we lost 3000 jobs that went to Mexico” than to specify that “we lost 3000 jobs when the factory moved to Mexico, however only half the jobs went to Mexicans, the other half is natural progress of technology and no one got those jobs”. It’s much easier to simplify and blame one thing, both for people in those circumstances, and even more so for the media who may want to push a particular narrative.
Similarly, there is a decent amount of evidence suggesting that perhaps places like China and other south-east asian countries were locations for equal or more jobs that got sent oversea. However, as before, I suspect that it is more likely that people simplify the message, and start using complaints about NAFTA as a short hand for complaints about all jobs lost to foreign countries, as its prominence in the news and being one single location next door, rather than lots of smaller deals over time with a number of far away places, makes it an easier target.
The number of jobs lost to NAFTA, or similar trade issues, or to include with automation, is quick tricky to pin down. One suggestion from the (surely completely unbiased) website ReplaceNafta.org uses figures from the government Trade Adjustment Assistence schemes to place a number of ~900,000 jobs lost from NAFTA and ~3,000,000 jobs lost from all trade. A paper from the Economic Policy Institute says that between 1993 and 2013, ~850,000 jobs were displaced. The important thing here is not to get bogged down in how many jobs precisely were lost and to where, but the over-arching narrative that a very large number of jobs were lost due to NAFTA, and I suspect a much larger number of lost jobs from surrounding circumstances were attributed to NAFTA than it deserved.
To briefly argue the other side, it is true that the loss of manufacturing jobs didn’t change rate with NAFTA, although it’s important to remember that NAFTA wasn’t a one-off implementation but had decades of smaller deals that led to it. Further statistical arguments can be delved into as to how one point on a flat line doesn’t mean it didn’t matter, but I think I’ll lose my whole audience if I go there.
What we need to focus on is the lasting impact on the overall narrative that these changes left. Today, only around 1 in 3 Americans think that NAFTA is beneficial, and “just 56 percent of Americans think international trade is on the whole good for the country.” With that established, let’s head back to 2016.
For Hillary Clinton, NAFTA has been a seagull around the neck for quite some time. The fact that it was signed by her husband perhaps put some unfair link between it and her in first place, but she did back it for quite some time. In 2008, Obama repeatedly whacked her with it, to her detriment in the primaries.
So far, we’ve only looked at the numbers around NAFTA, but I would highly recommend a brief detour to look at one town as an anecdote of the human effect of how it changed the course of the election. The piece is an interview with Lou Mavrakis, the mayor of a town that lost 2/3 of its population in the last few decades and the closure of its steel mill in the ’80s. He talks about how anyone publicly supporting Republicans in the past would get rocks through the window. He tried to get attention from the Democrats unsuccessfully to help rescue the town from its problems, but when Trump came there and promised a return of jobs, for the first time yard signs went up supporting the other side.
This is just one story, but it largely weaves in with what other people are saying. Presidential rising star Pete Buttigieg has said that the Democrats have ‘abandoned’ the midwest. The last Democrat to win in Michigan, Gary Peters also points to how failing to win “union members who didn’t trust her position on free trade” really hurt Hillary.
But from the starting position of struggling to get through to midwest voters on NAFTA, Clinton got whacked again on free trade when it came to the Trans Pacific Partnership. This time she got whacked from the left and the right. Bernie Sanders campaigned on opposing “job killing trade agreements like the TPP”, while in debates Trump won points by repeating that Hillary had called the TPP “the gold standard” in trade deals, which she then tried to get out off, which then got her a whole new round of attacks for flip-flopping.
So the theory is that opinions on trade seriously hindered Clinton in 2016 in the RB, but do the data back that theory up? It would seem so. According to RealClearPolitics, exit polling found a very clear divide when asking people whether (A) “international trade more likely takes away American jobs or (B) creates them”. Among voters in group A, Trump had a 34 point margin of victory (65 to 31). But that’s across the US – did people in the RB fall more into group A than B? From the piece:
“the gap between those who think that trade saps jobs and those who think it sustains them was much bigger in the Great Lakes region. Voters viewed trade as bad, rather than good, by 19 points (53 to 34 percent) in Pennsylvania, 16 points (48 to 32 percent) in Ohio, 15 points (50 to 35 percent) in Wisconsin, and 19 points (50 to 31 percent) in Michigan.”
So finally, let’s link this all back up and see if we have a convincing narrative. If you are a democrat from a region where either you, your family, or your town has seen a considerable loss of jobs in the last few decades, how would you feel when it comes to voting for a candidate who seems to have championed the policies that led to this situation, and seems to be suggesting even more deals to keep this going? Or how would you feel about the man from the other party to your normal vote who is championing killing the TPP, putting up tariffs and promising to bring back jobs? For some, perhaps they thought their best choice was to switch party and vote Republican for the first time in decades. For hundreds of thousands, it looks like they couldn’t stomach either option and simply stayed home on election day.
