Georgia: It Was Biden+Checks, Not Ossoff vs Perdue

Joe Biden and Jon Ossoff’s coalitions in Georgia on November 3 were highly similar but not exactly the same. Some people voted for Biden but not Ossoff. Some people voted for Ossoff but not Biden. In particular, many of the people in the first group were college-educated whites in places like North Atlanta, who generally favor Republicans but couldn’t stand Trump.

Despite generally being accurate in my predictions for the Georgia runoff, one thing I underestimated was the extent to which Ossoff’s underperformance in these hyper-educated suburbs would stick. In November, Biden won GA-06 (one of the most college-educated districts in the country) by 11 points, while Ossoff won it by just five. In January, Ossoff made ~3% of gains statewide, turning a loss into a win, yet he still only won GA-06 by five points. Perdue/Biden voters seemed to have generally stuck with Perdue. The logic behind this is fairly straightforward; these voters seem to generally want a bipartisan government, they’re not loyal Democrats and so the idea of a Democrat trifecta makes them a bit … nervous. And they already voted for Perdue once, why switch now? Especially when these voters are often rich and concerned about what Biden’s fiscal plans might do to their pocketbooks.

Essentially, instead of getting something closer to Biden’s more educated coalition, it appeared as if Ossoff won by sticking to his coalition but with some critical persuasion and higher Black turnout which gave him the boost he needed. What interests me though, is that the data suggests something different happened.

I used a linear regression model to analyse Ossoff January %, using Ossoff November %, Biden November %, white education rate, and % change in Black share of the electorate from November to January.

What these results showed consistently is that Ossoff’s January coalition can be more accurately described as “Biden’s November coalition, with educational depolarization” than “Ossoff’s November coalition”. In other words, the January runoffs saw polarization along Presidential lines and de-polarization along educational lines. Something that’s nearly impossible to see unless you use a technique like linear regression! In many ways it’s practically imperceptible from alternate hypotheses but the distinction is meaningful.

Why does this matter? Well, it seems important for analyzing what actually went down in January. Rather than seeing the resemblance between Ossoff’s two coalitions and saying “Ossoff voters stuck with Ossoff”, we can say that two countervailing forces were at work. On the one hand, we had Presidential polarization, by which Biden/Perdue voters were quite likely to switch to Ossoff, and Trump/Ossoff voters were quite likely to switch to Perdue. On the other hand, we had significant educational depolarization, where higher-educated voters became likelier to go for Perdue, irrespective of first round vote choice (and vice versa). I would say what happened is “people voted predominantly based on their opinion of Biden/Trump, but class interest reared its head due to the $1,400 stimulus checks on offer.” Coalition-wise the outcome is very similar but you can see how the analysis changes. Under the latter lens, we have more reason to believe that Ossoff and Warnock ran a strong working-class-focused campaign that was able to win lower-education voters by focusing on materially improving their lives with stimulus checks (among other things).

To look at it another way, picture a Trump/Ossoff voter, back in November. When we’re thinking about how they vote in January, we can choose to focus on either their support for Trump, or their support for Ossoff. One might be tempted to think that they’re an easy get for the Ossoff campaign, given that they already voted for him once. What this analysis suggests is that, if anything, these voters defaulted to their Presidential loyalty and were generally likelier to vote for Perdue. Many of them were persuaded to vote for Ossoff, but along class lines, presumably because of the kind of campaign Ossoff ran. Democrats should try to emulate his success in the future and promise poor people more free stuff.

NM-01 Is Not A Good Result for Democrats

If you’d been following media expectations ahead of the NM-01 Special Election last week, you’d probably think Democrats had a very good night there. Pundit Dave Wasserman said that Dems should be happy if they win by more than 15, and general expectations among those who follow this kind of thing seemed to place the median outcome at somewhere around D+17. Sabato’s Crystal Ball rated the race Likely D; implying a chance of Republican victory. When all the votes were counted, turnout was higher than for any other federal special election so far and Melanie Stansbury (D) prevailed by 25 points, beating Biden’s 23 point margin there. The problem? Media expectations were incredibly wrong. She should’ve won by more.

Two things led to people underestimating Democrat Melanie Stansbury in this special election. The first is now-Secretary of the Interior Deb Haaland’s poor performance in this district last November. She ran 7 points behind Biden, winning by a smaller 16-point margin. I think many people erroneously assumed that this was the baseline for a Democrat in this district but there’s no clear reason why – it’s far more likely that Haaland just wasn’t a very good candidate and Biden’s 23 point win is a better baseline. The second is the idea that we don’t know what turnout’s gonna be like. It is a special election, after all. In fact, we had a very good idea of what turnout was going to be in this election. Most of the vote was cast early and we knew what the breakdown was by party registration. Crucially, unlike in some other states, party registration tracks pretty closely to voting in New Mexico so we can infer quite a lot from this info. Based on this, it was pretty clear that Democratic turnout was strong. In the end, the electorate was D+23, compared to registered voters who are just D+19.

So, why does this mean the NM-01 result wasn’t really good for Democrats? Well, as far as I’m concerned, their target is holding the House. The relevant question is; was this election result in line with that outcome? On balance, I’d say not. Ultimately, Biden won this district by 23 points, but the electorate last Tuesday was probably Biden +27 or so. In other words, if you adjust for turnout then Stansbury actually under-performed Biden by two points. Two questions arise from this: should we be adjusting for turnout? And can Democrats afford to do two points worse than President Biden?

To me, the answer to the first is pretty clearly yes. It beggars belief that Democrats would have such a turnout advantage (nationally) in a Democratic midterm election, easily outstripping what they got in 2018. I think the clearer story here is that special elections are noisy and this is particularly true for turnout. Democrat turnout was good in NM-01, it was god-awful in TX-06, but our baseline expectation should be that 2022 turnout will be pretty similar to 2020, maybe a bit better for the GOP. The higher turnout an election is, the lower the likelihood of major turnout differences between the two parties anyway, as the pool of non-voters gets smaller. So yes, we should definitely be adjusting for turnout.

The answer to the second question is less clear, but probably not. On the one hand, the median House district is currently Lauren Underwood’s IL-14, which was Biden +2.4. So from that point of view, Democrats can afford to do two points worse than Biden. However with ruthless GOP gerrymanders surely on the way, it seems likely that the median district will shift to the right, perhaps by a point or more. And so Democrats will probably want to run as close to Biden’s performances as possible if they are to hold the House. Through this lens, while Stansbury’s vote margin is not strictly bad for the Democrats, it’s very hard to construe it as good.

NM-01 is the best data point we have so far for what the national political environment is like. Unfortunately, it tells us very little we didn’t already know. Democrats are probably ahead in the generic ballot, but probably not by enough to be favorites to hold the House. And most of all, there’s a lot of uncertainty. It’s useful to analyse these special elections but rarely a good idea to read too much into a single data point.

Civil Unrest in Kenosha Likely Helped Donald Trump

Introduction

The 2020 Presidential Election was one for the history books. A deadly pandemic, a uniquely unpopular President and mass protests stemming from the Black Lives Matter movement all culminated in an election result that was in many ways very expected (Biden wins by flipping WI/MI/PA + AZ/GA) and in many ways shocking (Biden takes huge losses with Hispanics, comes within 0.7% of losing the election and Democrats nearly lose the House). Perhaps nowhere symbolized the madness of 2020 more than the city of Kenosha, Wisconsin. In late August, a 29-year-old Black man named Jacob Blake was shot by a Kenosha police officer, sparking roughly a week of major civil unrest in the city, ultimately leading to the deaths of two more people. Property damage was estimated at up to $50 million.

One of the central questions of 2020 is to what extent did these tumultuous events affect the final election outcome. Was Trump helped or hindered by COVID and the civil unrest, did Democrats nearly lose the House because of “defund the police” or was it something else entirely? To a degree, these questions can never be fully answered. But that doesn’t mean we shouldn’t try. What follows is an attempt to use election and demographic data to determine the impact of the BLM protests/riots on support for then-President Trump. Given that I will predominantly focus on Kenosha, it may be hard to draw a meaningful conclusion for the entire country. That being said, I find strong evidence that the rioting in Kenosha resulted in increased support for Donald Trump, and that if we’d seen a similar level of rioting in say, Milwaukee, it might’ve cost Joe Biden the state.

Methodology

Say I were to claim that the Kenosha unrest helped Biden receive more votes in Kenosha City by increasing his support among Black and young voters. How can I test that claim? Or how could I test the opposite, maybe it hurt Biden with moderate/conservative white voters? I’d love to travel to an alternate dimension where the Kenosha riots never occurred and we could see what the difference in outcomes is. Unfortunately, this is clearly impossible. The question then becomes; how best to evaluate the counterfactual. One idea is that the default situation in any election is 50/50. Biden beat Trump in Kenosha City 56-42, so he clearly did well there, right? This idea doesn’t really hold up to scrutiny. Kenosha is a moderately diverse city in a moderately liberal state, it would be shocking if it didn’t vote Democratic. We need a different baseline. How about Hillary Clinton’s performance, four years ago? After all, the correlation between the two elections is very strong. She beat Trump 55-38 in Kenosha, three points better than Biden. Does that mean the riots cost Biden three points? Not necessarily. It’s possible that Kenosha’s demographics just weren’t very favorable to Democrats this year. Biden also did worse than Clinton in the neighboring city of Racine, were there were was far less unrest.

The solution is to use a method called linear regression; a model which can explain relationships between various inputs (say, race, ideology, education levels) and an output (Biden-Trump margin). By training this model on towns cities all over Wisconsin – anywhere over 15,000 eligible voters – we can get a very good picture of how we’d expect Kenosha to vote, then compare how it actually voted. It’s the best tool we have of determining what happened in Kenosha last November.

Results

The reason why Trump lost Wisconsin last November is pretty straight-forward; he lost ground with college-educated voters. In fact there’s a clear correlation between an area’s educational attainment rates and the 2016 -> 2020 swing against him.

The above graph shows the correlation using data from Wisconsin’s biggest towns and suburbs (excl. Kenosha). The interpretation of the trend line is that you’d expect Trump to improve by 2.1% in a town that is 0% college-educated, and you’d expect his vote margin to drop by 18.1% in a town that is 100% college-educated. A simple enough story. But what happens if we add Kenosha City and its suburbs (i.e. the other townships in Kenosha County)?

A shown in the graph, Biden under-performed in every single township in Kenosha County, relative to what we’d expect. This is not town-to-town variation, this is a systematic loss across the Kenosha area. In fact, of the ten townships, Biden improved on Clinton in only one of them. Suffice to say, if all of Wisconsin behaved like Kenosha then Trump would’ve won the state, perhaps easily.

You might also have noticed that the trend-line of the graph got steeper, after we added the Kenosha data. Or that if we drew a trend-line through the Kenosha data, it would be steeper than the previous data. In other words, it seems that Biden’s weakness in Kenosha was concentrated among lower-education voters and that he may not have suffered a penalty with college-educated Kenosha voters. Unfortunately, we do not have enough data points to properly test whether this is true – more granular (precinct) analysis is much trickier but will hopefully come at some point in the future.

Having analysed the data graphically, we can now try our hands at a linear regression. I assembled a variety of variables for this but it turns out only two proved useful: % college-educated and % 3rd-party. It seems that although Trump held on to nearly all of his non-college voters from 2016 as well as persuaded some non-college Clinton voters to flip Republican, he was doomed by 3rd party and college-ed voters breaking for Biden.

However, when we include the Kenosha data and institute a dummy variable for the Kenosha townships (1=Kenosha, 0=everything else), we find that Biden had a statistically significant under-performance there. The chances that this happened due to random variation are miniscule.

The regression model estimates that the “Kenosha” variable cost Biden 3.9% of margin, quite a lot for such a strongly-contested swing state. Given Kenosha County’s share of the Wisconsin electorate, this phenomenon could be said to have cost Biden ~3,500 votes statewide, a sixth of his winning margin.

When we restrict the linear regression to just Kenosha City itself and ignore the rest of Kenosha County, the estimated size of Biden’s loss gets bigger (4.8% vs 3.9%) but the effect gets less statistically significant as we now only have one data point to work with. This is consistent with the idea that Kenosha City was the source of Biden’s weakness, and that the effect spread out with decreasing strength as we get further away from the city into the suburbs and nearby rural areas.

Establishing Causation

One thing is entirely clear from this data; Joe Biden did several points worse in Kenosha than he should have. This fact alone need not command our attention were it not for two things. Firstly: we live in an age where counties rarely vote very differently from Presidential election to Presidential election, and when they do move it is often for clear demographic reasons (see: Rio Grande Valley for Trump, or Atlanta suburbs for Biden). When a city does not behave as it ought to, it is useful to go looking for answers. Secondly: the civil unrest that happened in Kenosha last year was one-of-a-kind, as I laid out in the introduction. No other city of a comparable sized was affected by the Black Lives Matter protesting as much as Kenosha was.

The question then becomes, how do we prove (or disprove) the idea that the rioting in Kenosha hurt Joe Biden? In a truly rigorous sense, we can’t. We can’t know 100% for sure what happened in Kenosha last year. However I’ll accept a lower burden of proof than 100% and having approached this analysis without agenda I am now thoroughly convinced that the Kenosha unrest hurt Biden’s vote share there last November. There are a few reasons why. Firstly, we know something happened in Kenosha to cause a statistically significant and unexplained increase for Trump there and I have been unable to find any alternative explanations that stand up to scrutiny. Secondly, there is decent corroborating evidence for what we’ve observed here:

Image

This graph, courtesy of David Shor, shows how the George Floyd protests majorly raised the salience of policing as an issue nationally, and it’s reasonable to think this jump would’ve been even bigger in Kenosha, which saw much more protesting per capita than your average city. The key things to understand here are that in most places in America, including somewhere like Kenosha, the median voter is pretty pro-police, including a lot of Clinton voters. Polarizing an election along this issue is therefore bad for Democrats. It’s also true that Clinton voters who are strongly supportive of the police are disproportionately non-college-educated, so we can expect any losses from the salience of policing to hit Democrats particularly hard with this group. This is exactly what we observed in Kenosha.

We also have research from academic Omar Wasow which suggests something similar happened nationwide in 1968. He found that “violent protests likely caused a 1.6-7.9% shift among whites towards Nixon’s ‘law & order’ campaign and helped tipped the election.”

Image

In particular, violent unrest may well have cost Democrat Hubert Humphrey the critical states of Missouri, Illinois, Ohio, New Jersey and Delaware. Wasow does an incredible job of establishing causation for this fact, using 1968 rainfall data as a way of randomly simulating which cities see more violent protests than others. It’s a really interesting study, I fully recommend it. It also corroborates the over-arching theme of this analysis of Kenosha, which is that voters generally like the police and “law and order” and when they feel threatened by civil unrest they will gravitate towards the party seen to be stronger on those issues.

Conclusion & Importance

Having examined the available and relevant data, my conclusion, which may not hold up as a PhD thesis but which I firmly believe is supported by the evidence, is that the civil unrest in Kenosha cost Biden somewhere in the range of 2-5% of vote margin in the city itself, and an unknown amount of vote margin in the surrounding areas and indeed across Wisconsin and the whole country. It also seems likely that these losses were mostly from voters who have not graduated from college although I would love to try and analyse that further using ward/precinct data from the city.

If the protesting in Kenosha was a stand-alone event, we could be forgiven for dismissing its importance. After all, it’s just a few thousand votes. But given the widespread protesting and rioting that America saw for weeks during 2020, it is incredibly important that we understand what happened in Kenosha so that we can try to extrapolate it nationwide. People will argue for years about whether BLM nearly cost Biden the election, but Kenosha offers us a rare opportunity to try and answer that question without relying on potentially flawed polling data. Trying to draw a conclusion about a possible counterfactual where George Floyd isn’t killed and all of these events don’t come to pass is a fool’s errand, but given the electoral outcome of the unrest in Kenosha (plus the available polling data) I think it is reasonable to conclude that Biden might’ve done a point or two better in vote margin nationwide had it not been for the Black Lives Matter protests.

The purpose of this article is not to tell people they should never protest nor is it for me, a white person, to tell Black people they need to tone down their outrage at their mistreatment. It is to try and build a better understanding of the electoral consequences of protests like these, as it is at the ballot box where real change occurs. Personally, it deeply concerns me that a sizable number of people who probably voted for Clinton and would’ve otherwise supported Biden ended up voting to re-elect Donald Trump because of Black Lives Matter. But such is the world we live in.

How Troy Carter Won in LA-02

Yesterday Troy Carter (D) beat Karen Carter Peterson (D) to become the new representative of the majority-Black Louisiana’s 2nd congressional district. This outcome shouldn’t surprise many, given that Carter is an established local politician with plenty of endorsements, and the fact that he was the more moderate candidate in a district that does not have strong demographics for progressives nationally. What does come as something of a surprise to some, including me, is the margin by which he defeated Karen Carter Peterson (KCP). Carter beat her by double digits (10.4% margin of victory), even winning Orleans Parish, easily the most progressive parish in the district. How did he do it? Here’s a few quick answers, with some graphs provided

1. Republicans voted nearly unanimously for Troy Carter

The biggest hurdle for KCP was always this: 23% of the district voted for Donald Trump, and those Trump voters can vote in this Democrat vs Democrat runoff. An examination of the data through linear regression shows that, of the ~16.5% of voters who picked a Republican in the first round and then voted in the runoff, roughly 90% backed Troy Carter. This, as well as the Republicans who strategically backed Carter in the first round (and presumably stuck with him for the second round). Overall it’s clear that if Trump voters are removed from the equation, Karen Carter Peterson would be heading to Congress. Very grim.

2. KCP lost a lot of ground in Orleans and it’s not clear why

Orleans Parish, the most progressive parish in the district (and the only one where white voters skew Democrat) makes up roughly half the vote. Outside of Orleans, KCP generally did what she needed to win the runoff. Orleans though, was a total shitshow for her. If I run a regression across the precincts and early voting totals to determine what factors were predictive of a KCP win, we get a picture that largely makes sense. KCP+Chambers voters went overwhelmingly for KCP, Carter+GOP voters were strongly for Carter. What doesn’t make sense is the fact that, after adjusting for other factors, KCP did 8.5% worse (17% of margin) worse in Orleans than the rest of the district. I have no idea why this happened, especially given she was endorsed by the New Orleans mayor, and I’d love to hear some theories. Note: it can’t just be that Chambers voters in Orleans didn’t turn out for her, because Chambers first round % is a variable I’m controlling for. Overall, if KCP had done as well in Orleans as she did in the rest of the district (after controlling for confounding variables), she would’ve lost by just 1-2%.

3. Progressives voted for KCP but turnout was weak

Based on the regression model above, I estimate that over 90% of Chambers voters who turned out in the second round voted for KCP. In isolation this stat is very good for her, I certainly would’ve been pleased if I’d heard this before the results came in. The trouble is that Chambers voters did not turn out at the same rate as other Democrats; to be expected although one might’ve hoped otherwise given Chambers’ huge effort to get people out to vote for KCP in the runoff. Overall I estimate that if Chambers voters had turned out for the runoff at the same rate as KCP/Carter voters then it would’ve cut Troy Carter’s margin from 10.4% to 2.6%. That is, assuming that the Chambers voters who stayed home aren’t too ideologically different to those who voted in the runoff. In reality, Chambers voters turned out at roughly the same rate as GOP voters, which is pretty bad given it was a Democrat vs Democrat runoff.

Which Safe Republican States Are The Least Safe

With recent high-quality polling showing President Trump trailing Joe Biden by double digits, the big electoral question on our minds these days should not be “who’s going to win?” but rather “how much is Biden going to win by?”

As a result, there has been some speculation as to the states currently rated “Safe Republican” that could hypothetically vote blue for President this November. Recent internal polling apparently shows Trump trailing in Kansas and only slightly ahead in Montana, two states he won by twenty points in 2016. These are far from Biden’s only reach options in a landslide scenario either; his campaign could also, in some plausible universe, target Alaska, Utah or South Carolina. The following is a ranking of which of these states are likeliest to go for Biden, from least to most. If you don’t want to read my full explanations, just take a quick look at this table which summarises Biden’s chances of winning each state in terms of a few key attributes.

Screen Shot 2020-07-10 at 2.34.20 AM

Alaska gets an “A” rating on the question of 3rd party voters, for example, because it had a lot of them in 2016 and Trump is unlikely to do well with them this time around. Meanwhile Montana gets an “F” rating on population density because it’s really really rural.

5. Montana

Despite being thought of as perhaps the likeliest state on this list to go blue, Montana really isn’t particularly primed to vote for Joe Biden. Why do people like Kyle Kondik think otherwise? My guess is a kind of recency bias. Montana is the only state on this list that has been competitive in the last 20 years; Obama came quite close to winning it in 2008. When it comes to a record of voting Republican for President, Utah, Kansas, South Carolina and Alaska are pretty gosh darn reliable. Or maybe it’s that incumbent Senator Jon Tester (D-MT) won re-election last year, even though Rosendale was, by all accounts, a sub-par candidate and no one’s suggesting that Joe Manchin’s similarly-sized win in West Virginia makes the state competitive at a Presidential level. Anyway, Montana has very few of the components that Trump defeat possible.

According to polls and recent election results, a lot of Trump’s losses from 2016 are going to come from college-educated whites. 33% of Montana’s white population has a college degree, compared to 35% nationally – not super ideal. Urban areas are also a liability for Trump, where even Republicans are more socially liberal. However Montana is the second-most rural state in America, behind only Wyoming. Nor does Montana have a rapidly diversifying electorate like one could say of states like Arizona, Texas or Georgia. Finally, and most importantly, Trump got 55.7% of the vote in Montana in 2016, second only to Kansas on this measure.

The main asset the Biden camp has in Montana is the fact that there’s a tight race for Senate going on at the moment thanks to the candidacy of Democratic Gov. Steve Bullock. This will certainly mean more investment in the state than Democrats would otherwise have seen.

4. South Carolina

In a sense, South Carolina should be the easiest state for Biden to win. Hillary Clinton got 40.7% of the vote here, far more than Alaska (36.6%), Kansas (35.7%), Montana (35.4%) or Utah (27.2%). Yet this obscures an important feature of South Carolina politics; rural white evangelicals make up a large portion of the state’s Republicans and most of them will never, ever, vote for a Democrat. I call this state of being ‘pure insanity’ and to me there are two main warning signs; a lack of third-party voting in 2016 and a lack of swing in elections where the GOP candidate is god-awful. The former demonstrates a willingness to recognise that just because Clinton is bad doesn’t mean one has to vote for Trump and South Carolina had one of the lowest rates of third-party voting in the country in 2016. Similarly, the latter demonstrates an inability to recognise the flaws of candidates belonging to your party and/or an overarching belief in conservative ideology above all else. We have no such election results from South Carolina to point to but it’s relatively similar to Alabama and I hope we all remember how many rural evangelical whites in Alabama happily marched to the polls to vote for far-right pedophile Roy Moore.

Apart from this, South Carolina does have a few things going for it. With an electorate that is 27% black, South Carolina is far better for Democrats on a racial level than any other state on this list. I would be highly surprised if black voters in South Carolina gave Joe Biden less than 90% of their vote. Additionally, it isn’t super rural; on a level with Kansas but far less so than Montana or Alaska.

Overall, although it is the closest state on the list (Trump +14), Biden may have a hard time winning simply because of the ‘pure insanity’ of the state’s white population.

3. Kansas

Of the five states on this list, Kansas is the only one where Trump underperformed Romney; it swung 1.17% Democrat from 2012 to 2016. That alone merits consideration as a state that can possibly vote Democrat this cycle. No other state which voted for Trump by a similar or larger margin swung left in 2016. Kansas does have a fair few demographic features going for it as well. At 35.4%, its white population is slightly more college educated than the nation as a whole. It’s also at a similar level of population density as South Carolina – not particularly high, but far above Montana or Alaska. Kansas is perhaps the most ancestrally Republican state on the list, which may seem like a downside but I view as an upside – it means that there were some voters who cast ballots for Trump in 2016 simply because they were used to voting Republican.

Demographically, Kansas has some of the best ingredients for moving left this cycle and I have no doubt Trump will do far worse here than in 2016, but at the end of the day he has two main advantages. Firstly, he got 56.2% of the vote here, which is really difficult to overcome. Secondly, it simply doesn’t have a high-variance/low-information demographic like Utah or Alaska do. More on this in the following sections.

2. Utah

Demographically, Utah shouldn’t be as Republican as it is. It’s pretty white, sure, but it’s also reasonably urban – far more so than any other state on this list – and its white population is the most college-educated of the five, at 36%. The main reason why it’s been so solidly Republican in recent years is religion; most Utahns are Mormon. In 2016, Trump’s unique weakness among Mormons and the candidacy of Evan McMullin largely negated this effect and he won Utah by 17 points, a result much more in line with the state’s demographics.

As I see it, there are two ways of looking at this state. One is that Trump got just 45% of the vote here and most people who didn’t vote for him in 2016 strongly dislike him. Given his poor standing in national polls, that makes him far from a lock to get the ~50% he needs to win. The other is that Clinton got just 27% of the vote, much less than the 34% Obama got in 2008 (I ignored 2012 because Romney was Mormon). In that sense, the 2016 results were even further proof that while Mormons may strongly dislike Trump, they are far from willing to vote for a liberal Democrat. Through this lens, Biden’s climb to 50% is nearly impossible.

As is often the case, I think the truth is somewhere in the middle. If we view 3rd party voters as a homogenous bloc, Trump has big problems in Utah. Yet it is simultaneously true that McMullin voters are overwhelmingly conservative and unlikely to break for Biden in the same large numbers that say, Stein voters in California might. On this basis, if I had to guess then I would say Trump wins Utah by a bigger margin than Kansas or South Carolina. The main thing going for Utah as a potentially competitive state is variation. We simply know a lot more about voters in most other states; they are much more easily sorted into ideological and demographic categories. Mormons are a big question mark, and so that increases the ‘probability’ that Utah votes blue even as Biden’s expected vote share is relatively unchanged. If I wake up on November 4th and Trump has won Utah by 16 points or lost it by 3 points, I wouldn’t be too surprised. I can’t say the same for a state like South Carolina.

1. Alaska

Alaska is certainly the state currently rated Safe R in which I am most optimistic about Biden’s chances. There’s a reasonable amount going for him there. Its white population is relatively college-educated, at 36%. It is highly rural, yet white voters are disproportionately located in more densely populated areas. Most importantly, it gave Trump just 51% of the vote in 2016 – less than he got in Ohio! Unlike Utah there’s no indication that third-party voters, who were more numerous in Alaska than almost any other state (by percentage), are particularly conservative. Even if Trump keeps the vast majority of his supporters from 2016, Joe Biden could theoretically win Alaska’s three electoral votes by winning 2016 third-party voters overwhelmingly as well as voters who didn’t cast ballots in 2016.

ME-02?

It’s worth mentioning Maine’s 2nd congressional district, which is currently rated Lean/Likely R by election forecasters but which could easily stay red even if states such as Kansas or Alaska go blue. To compare to Kansas for example, ME-02 is far less educated, far more rural, more white and, by available metrics, only slightly more liberal. It did have a decent amount of third-party voting in 2016 and will see Ranked Choice-Voting used this November and for this reason I am in two minds as to whether it should be rated Likely or Safe R. But if I had to place ME-02 on this list, I would say it is probably more secure for Trump than Alaska.

 

 

 

 

Why We Vote

I’m increasingly convinced that being ahead of the curve in terms of macro-political predictions doesn’t just mean having the best datasets or paying closest attention to the polls or having the best model. If you want to see things in advance, you’ve often got to be able to re-examine commonly held axioms about how elections and politics work on a fundamental level. These re-examinations will come regardless, in time, following new polling data or actual voting results but that doesn’t mean they can’t be made earlier.

My theory of the day relates to the fundamental reasons why people vote. I’m of the view that the only sustainable long-term reasons for voting behavior are identity and ideology, and the main ideological questions that really matter during this current era are culture war ones. In essence, voting becomes a simple function of a few inputs, such as the voter’s race, stance on guns, stance on immigration, stance on racial issues, stance on LGBT issues.

When pundits say “partisanship is increasing”, it’s true if you take that to mean down-ballot elections are more and more closely resembling presidential elections. What I’d say, though, is that elections are increasingly driven by ideology – and a narrow range of ideological questions at that. This happens foremost on the Presidential level and less so once you get down to local offices, but it is taking place everywhere. Some measures of partisanship, such as voter registration, are actually increasingly decoupled from federal and statewide elections. In most of rural, white Louisiana, for example, Trump won registered Democrats by a landslide. His performance more closely mirrored previous referendums Louisiana had passed such as a vote to ban gay marriage and a vote to protect the right to bear arms,

Before we look at the data, let’s try and explain why this is. Basically, a lot of voters will pick a Democratic candidate for a given office simply because they are some combination of a) registered as a Democrat, b) identify as a Democrat or c) have voted Democrat a lot previously. This is increasingly less true, but it’s still a major predictive factor and I believe it to be on its last legs. The reasoning is pretty simple – in this age of mass media and nationalized politics where parties will consistently try and sell themselves to voters on a number of culture war issues like guns or immigration, it’s increasingly difficult for a voter who is generally liberal on these issues to vote Republican, or vice versa. This doesn’t necessarily happen overnight, either. A culturally conservative voter might’ve voted for Obama in 2012 out of party loyalty/identity and then stuck with Clinton despite some reservations for the same reason. That voter is far from a sure bet for either party in 2020.

In other words, saying a certain district or state voted for Clinton or Trump by a certain amount and will therefore vote a certain way in 2020 is, in my humble opinion, quite a backwards way to look at things. Presidential votes are mere outputs of the election function, if you try and predict the next output using the last one you’ll always be one cycle behind. The inputs – the reasons why voters vote they way they do – are where the real insight can be gleaned. For example, why is Washington guaranteed to vote for Biden this November? Not because it voted for Clinton, but because it’s a reasonably urban state that, despite being whiter than average, is highly liberal on issues such as gay marriage and abortion. 

Now it’s time for some evidence, so far I’ve just been theorizing. For a moment let’s take ideological sorting out of the picture and look at the basic concept of inputs vs outputs.

PoliticalKiwi2

If I’m right, historically speaking, when a state’s inputs are at odds with its outputs, there should be some kind of correction in subsequent elections. Perhaps unsurprisingly, there are a number of such occurrences. The most obvious ones are home state effects, although this idea is more commonly held. It’s not controversial to suggest that South Dakota trended right in 1976 because favorite son George McGovern was no longer on the ticket. There are other examples, however, which lack such easy explanations. Take Mississippi in 1976. With my approach, it doesn’t make much sense for it to have voted 11 points to the right of Alabama. After all, Mississippi has a higher black population, it’s only a little bit more conservative and it’s less urban (rural whites used to be the most Democratic in the South). Sure enough, in 1980, they voted exactly the same (to within 0.02%!). Or we have the example of Tennessee in 1984. Reagan won the South in a landslide that year yet, for lack of a clear reason, he did about six points worse in Tennessee than any other Southern state. Sure enough, because it was already more Democratic than its inputs would suggest, Tennessee was the only state not to move left in 1988.

PoliticalKiwi3If you looked at this map you could be forgiven for assuming that Republicans nominated a Tennessean in 1988, or at least that they campaigned there extensively and successfully. In reality, however, this is a state that’s out of wack realigning itself.

An example you might be more familiar with would be Indiana, in 2008. Obama pulled off an impressive win there, beating McCain out by 1% four years after Kerry had lost to Bush by 20%. This is often attributed to Obama being from neighboring Illinois and his superior ground game, but I’m not sure that’s really what’s at work here. After all, Illinois itself didn’t move left anywhere near as much as Indiana. Nor did, Ohio, a strongly contested swing state with similar demographics. What my theory suggests is that in 2004 Indiana was far more Republican than it had any right to be, and 2008 was a correction to that fact. To see whether this is true, I looked at Missouri, a state that is incredibly demographically similar to Indiana and has voted in lockstep with it since 2008. In 2004, Bush won Missouri by seven points, while winning Indiana by 20. If you’d used my theory at the time you might have wondered why Bush did so well in Indiana despite all the relevant inputs, and concluded that Democrats had a good deal of potential upside there in future elections.

Looking at historical examples on a state level is all very well and good but now it’s time to bring ideology back to the forefront of the theory and look at how that has played out in recent years. I’ve decided to focus on a single state (Maine) so that I can make apples-to-apples comparisons using referendum data. Why Maine? You may have heard a lot about it from me recently. Three main reasons; it has a wealth of useful referendum data, it’s overwhelmingly white (which allows me to essentially control for race) and it has a number of small townships which enables me to do granular analysis without going to a precinct level. A disclaimer: I could’ve done a similar analysis on a number of states and gotten very similar results, they are by no means unique to this one state.

In 2012, Obama won Maine by a landslide. With a linear regression model we can gain some insight into the factors that predicted support for him. In this case, Obama’s win was overwhelmingly a function of party registration, not ideology.

PoliticalKiwi4

Being more liberal on certain social issues was associated with voting for Obama, notably gay marriage (associated with lack of religiosity) as well as support for Medicaid expansion and opposition to religious exemptions for vaccination requirements. At the same time, however, support for gun control (associated with pop. density) actually made voters less likely to support the President, as did having a four-year college degree. These results speak to Obama’s strength with secular, rural, white working-class voters that we saw across the Midwest and North-East in 2012. They also speak to how little of Obama’s big win in Maine was built on an ideological bedrock, making it inherently unstable.

From the above table alone, we can see two major warning signs for Democrats in Maine prior to the 2016 election. Firstly, a large number of people voted for Obama simply because they were a registered Democrat (or rather, because of the behaviors that being a registered Democrat implies). This is not a sustainable reason to vote Democrat in the long-term; voters will generally gravitate towards the party that represents them on the issues of the day and these trends are more powerful than ever thanks to mass-media and the prominence of the ‘culture war’. Secondly, the issue that drove voting behavior the most in 2012 – gay marriage – both uniquely benefits Democrats in secular Maine and is likely to fade from the public consciousness as conservatives gradually accept defeat on this issue. We know how the story goes from here; Trump does quite well in Maine and becomes the best-performing Republican there since George H.W. Bush. The next table shows the change between 2012-2016 regressed across the same variables plus the 2012 pres variable:

PoliticalKiwi5

Interesting to note that once you account for support for Obama in 2012, every single other variable in the dataset is associated with a swing to Clinton in 2016, even party registration! This too makes sense intuitively; if we hold ideology constant then it’s entirely plausible that a Romney-voting registered Democrat has an above-average chance to flip to Clinton, maybe something about Romney or Obama as candidates appealed to them or their specific area. The results of this regression generally confirm that it’s not really about party registration, although that is one of my preferred ways of expressing (historical) partisanship. The point is that where a region votes in a way that is inconsistent with their ideology across relevant issues, they are likely to trend in that direction. Finally, let’s look at a regression of the 2016 presidential results using all the variables above (i.e. the original set plus 2012 pres):

PoliticalKiwi8

These results show that while Clinton broadly did well with voters who were ideologically liberal, there was still a great deal of variation in the results that could only be explained by previous partisanship, either through registration or through voting Obama in 2012. In other words, although it may seem somewhat implausible, there were a lot of conservative Clinton voters and liberal Trump voters in Maine (and likely in the rest of the country too). Important to note here that I don’t mean whether or not a voter would describe themselves as ‘liberal’ or ‘conservative’ in a poll; a flawed metric that doesn’t really capture their true ideological bent. What I’m saying is that in Northern Maine for example, there were a non-trivial number of Clinton voters who do not support gay marriage, do not support gun control and do not support medicaid expansion.

(As an aside: it’s worth noting that many of these variables come from different years i.e. party registration from 2020, gay marriage from 2012, medicaid expansion from 2017, however other research I’ve conducted suggests that this will have little overall impact on the results of my regressions.)

Let’s zoom out and look at the big picture now. What am I getting at here? Firstly, as I’ve shown, it pays to view elections (at least on the Presidential level, where most of my analysis has focused) as a function with inputs and outputs, and we must not confuse the two. Historically there have been a lot of inputs that differ from the core inputs of today (race & ideology) yet it still held true that where there was tension between those inputs and outputs, future election results would gravitate towards the former. Secondly, aside from race (there’s a lot of research to be done into how race and ideology are intertwined but we can leave that for another day), it is increasingly true that the only relevant input is ideology, specifically social issues. If this weren’t true then the first point would largely be moot as there’s no sense basing a theory of elections around these mythical ‘inputs’ if we have no idea what those inputs actually look like. This idea has been shown empirically in the difference between the 2012 and 2016 Presidential results and I believe it also makes sense intuitively. The nation has been heavily divided on social issues before (if people try and tell you that America is more divided than it’s ever been, remind them that The Civil War Happened) but the age of mass-media has enabled a nationalization of politics like we’ve never seen before. I would also argue that voters are likelier to vote based on ideology when one party tries to focus an election on that specific issue (i.e. the GOP and immigration). Just look at what happened in 1964 election, where a large number of Democrats voted for Barry Goldwater and an even larger number of Republicans voted for Lyndon Johnson on the basis of their stances on the Civil Rights Act.

Furthermore, I’m generally skeptical of claims that 2016 trends will be nonexistent or even reversed in 2020, on the basis that the underlying reasons have, if anything, gotten stronger. Ask yourself, has the GOP tried to distance itself from the Trump campaign’s messaging in 2016, or has it doubled down? You could also think about the incentives that act on the Trump campaign this time around. Trump is personally unpopular and will continue to be at least until he leaves the White House, so their best hope of winning in November is by polarizing the electorate along ideological lines and convincing culturally conservative voters that although they might not like Trump, he’s better than what the Democrats offer. This strategy also pays dividends in terms of the Electoral College/popular vote split. As for the Biden camp, they’ve signaled they want to make this election a referendum on Trump (they will clearly win if they do, the man is unpopular). How this relates to ideology as we’ve examined it here is perhaps unclear, however I do know that if the Trump campaign wants to put certain issues front-and-center, Biden will have a hard time preventing it.

The data is clear on the current trend of ideology playing a larger role in our Presidential elections and I feel reasonably confident in saying that trend will continue in November. What this will look like will ultimately depend on the specific issues that come into play (the COVID-19 pandemic is something of a wild card, for example). Broadly speaking, though, it means that Obama-Trump voter from Wisconsin who thinks illegal immigrants should be deported probably isn’t coming back to the Democratic Party.

 

 

Don’t Just Look At Competitive Elections

One of the big themes of the 2018 midterms, one of the core narratives that emerged from that day was the Democratic Party undoing a lot of the damage Trump had done to their coalition across the Rust Belt. From Iowa to Pennsylvania to Maine, Democratic candidates for House, Senate and Governor won back rural white working-class voters who had cast their ballots for the President in 2016. Sure they often couldn’t quite match Obama’s numbers, but they didn’t need to with their newfound gains in diverse states like California.

It’s a fine narrative, don’t get me wrong. There’s a lot of hard evidence there and some decent conclusions to be drawn. That being said, I wonder if our tendency to ignore races which went for either party in a landslide precisely because they were non-competitive blinds us to certain trends which emerge in these elections. Given how much down-ballot voting is now driven by partisanship and views on the President, there’s no reason we can’t look at R+30 districts for future electoral indicators just as we look at races which were decided by just a few thousand votes.

As a function of their competitiveness, certain regions and demographics tend to go underrepresented in the closest races. In the 2018 House elections, it was the South that saw less than its fair share of the battleground. In particular, as a function of gerrymandering and poor investment strategies by the DCCC, many urban and suburban areas across the South, areas with high rates of college education, did not see competitive House races. Birmingham, Memphis, Jackson, Nashville, New Orleans, Jacksonville, Tallahassee, Huntsville, Wilmington, Columbia and Baton Rouge are just some of the cities in the South located entirely within seats rated as Safe Republican or Democrat.

As for the races that did end up being close? In many cases, Democrats exceeded expectations in the suburbs. In Texas, they notched bigger-than-expected wins in the 7th and 32nd districts, while coming close in districts such as the 10th, 21st and 24th which weren’t considered particularly competitive. In Georgia, Republicans were generally acknowledged as having the advantage in the 6th district after their 2017 special election win, only to see it flip blue. The 7th district was considered even more of a reach for Democrats, and it ended up re-electing its Republican congressman by a mere 400 votes. Oklahoma’s 5th and South Carolina’s 1st (Oklahoma City and Charleston respectively) were two of just four districts rated as Lean or Likely Republican that voted Democrat in 2018 and it’s no coincidence they are both urban/suburban districts with reasonably high education levels. The other two upsets were a Clinton +16 district (California’s 21st) and the densest Republican district in the country (New York’s 11th – Staten Island and Brooklyn).

With this data in mind, a few days ago I posed a question. We’ve seen that Democrats generally did well in competitive races across the suburban/educated South but how did demographically similar areas vote in “Safe R” races? Surprisingly Democratic, comes the answer.

Not only are there some districts in the South where 2018 Democratic candidates for Congress improved on Hillary’s 2016 performance in cities/suburbs/college towns *relative* to rural areas, this proved to be the rule rather than the exception. This trend was particularly prevalent around the cities of Atlanta and Austin where Stacey Abrams and Beto O’Rourke proved successful in turning out Democratic-leaning voters at higher rates than your usual midterm. Districts such as Texas’s 17th and Georgia’s 3rd are largely white and rural and so remained safely Republican and will continue to do so, yet the biggest 2016-2018 swings occurred in the dense and educated areas of those districts; places which had already seen sizeable swings to the left from 2012-2016. Yet in Alabama without an inspiring figure at the top of the ticket, the widening education gap once again reared its head. Rep. Gary Palmer (R- AL06) managed to do roughly six points worse than Trump’s 2016 performance in the Birmingham suburbs (Jefferson and Shelby counties) yet was able to match the President’s performance in the surrounding rural areas. I could name numerous other examples but the point is that this phenomenon is widespread. It’s not unique to the South either but it’s prevalent there because it’s got the highest concentration of college-educated/suburban Republicans in the country.

The moral here is, if all you do is look at the competitive races the media reports on (with good reason, they decide the control of government), you’re not getting the full picture. In the case of 2018, a major missing puzzle piece is Southern metropolitan areas continuing their leftward leap. Of course, no trend is set in stone years before an election actually takes place. However perhaps they’re more set-in-stone than they  used to be and so I’m reasonably confident when I say that the results of the 2018 midterms are decent evidence that Trump’s biggest liability is, and will continue to be, college-educated voters in the South.

An incomplete list of Republican-held districts which saw a widening of the urban/rural divide in 2018, as compared to 2016:

Alabama 1st, 3rd, 5th and 6th, Arkansas 1st and 3rd, Colorado 4th, Florida 3rd, Georgia 1st, 3rd, 9th, 10th, 12th and 14th, Missouri 4th and 7th, Mississippi 1st, North Carolina 6th, 7th, 8th, 10th and 11th, North Dakota at-large, Nebraska 1st, Oklahoma 1st and 4th, Oregon 2nd, Pennsylvania 12th, South Carolina 2nd and 5th, Tennessee 4th, Texas 1st, 5th, 8th, 10th, 17th, 25th and 36th, Virginia 6th, West Virginia 1st. 

 

 

An Unusual Generational Divide In The Secular North

As a general rule, Democrats do better with younger voters. If the voting age was 40 instead of 18, Trump would’ve won the 2016 election in a landslide. It might also not surprise you to know that this is largely a function of factors such as ideology, population density and race. Youth don’t tend to vote Democrat simply because they are young, they vote Democrat because they are likelier to live in cities, to be non-white, and to agree with the Democratic Party on a variety of social issues.

Texas, New Mexico’s 2nd congressional district and Kansas’s 2nd congressional district are three places polled by the New York Times in 2018 where support for social issues such as the right of NFL players to kneel, an assault weapons ban or the border wall was also polled. In each of these places, after controlling for race, stance on social issues, party registration (where possible), education and region/pop. density, age had next-to-no effect on a voter’s opinion of Trump. This seems fairly intuitive, after all why would a 26-year-old non-college-educated white social conservative differ significantly from a 68-year-old non-college-educated white social conservative?

According to the same New York Times data, there is a region of the country where this pattern doesn’t hold up. In rural, white working-class areas of the North-East and Upper Midwest, young voters are significantly likelier to support Trump after controlling for all relevant factors in the data set. In Minnesota’s 8th district, being under 50 was associated with a 12% jump in Trump’s approval rating, compared to those over 50. In Maine’s 2nd district, being under 35 was associated with an 11% jump in Trump’s approval rating. These effects are statistically significant and if it doesn’t sound like much, it’s about equivalent to the difference between New Jersey and Ohio.

This polling dataset from 2018 isn’t the only place this phenomenon shows up, either. Hillary Clinton won youth by a wide margin in 2016 according to (admittedly flawed) exit polls, however in Maine the correlation was reversed and Trump actually performed best with younger voters. It was the over-65 group who gave Clinton the widest margin.

One might now be tempted to ask, why are young people in this area of the country so goddamn conservative? What is it about this cohort that finds a political home in Trumpism? I think, however, that is the wrong question. Youth in Minnesota, Maine, Michigan’s Upper Peninsula, upstate New York, wherever, are all exactly as Trump-friendly as they should be, based on their demographics. They are not very college-educated, they are overwhelmingly white, they tend to live in rural areas. It’s no surprise Trump is over-performing here in his approval ratings.

In actuality, it’s the elderly voters in the ‘secular north’ who are the outliers. Your average 65-year-old voter in ME-02 or MN-08  is white, doesn’t have a college degree, lives in a small town, is reasonably conservative, yet still disapproves of Trump’s job as President. Why this is the case is an open question but my working theory is that it’s due to a mish-mash of cultural factors dating back decades such as union strength in Minnesota’s Iron Range and French Catholicism in parts of Maine. My working theory is also that these factors are becoming rapidly less relevant and it’s part of why the secular north swung so dramatically from Obama to Trump.

What the New York Times data conclusively shows is that Trump has a lot more upside in these places than the conventional wisdom may assume. In fact, to be highly competitive in a state like Maine, all Trump has to do is convince a bunch of elderly voters who fit perfectly into his target demographic that he’s on their side. For sure, they may stick with their historical allegiance and continue to snub him. But Trump’s messaging seemed to work incredibly well in this area in 2016 and given how he’s only doubled-down since then, I like his chances.

 

 

Trump’s Big Southern White Voter Problem

You’ve probably heard it before. In 2016 Donald Trump had a historically poor performance with college-educated whites, allowing Hillary Clinton win Republican-leaning suburbs in places like Colorado and Virginia. However these were not the states that proved to be decisive in the Electoral College – it was less educated states such as Wisconsin and Pennsylvania that got the President to the magic 270 number while losing nationally by 3 million votes.

True to form, if you look at white education rates by state there is something of a pattern. Clinton excelled in higher education states such as California (42.5%) and Maryland (43.9%) while getting crushed in areas where whites tend to have fewer college degrees such as North Dakota (29.9%) and West Virginia (19.6%). On closer look though, that’s not really the full picture. In a lot of states, education doesn’t capture what’s actually going on. Why, for example, did Clinton do roughly 12 points worse than Obama in Maine, where 30% of whites have college degrees, yet improve on Obama’s performance in Louisiana, a state where just 27% of whites have college degrees? We can go even further. Connecticut is one of the most college-educated states in the country, Arkansas one of the least. Hillary Clinton lost more white support relative to Obama in the former than in the latter. It doesn’t add up.

One explanation that springs to mind is that education just wasn’t the factor we thought it was. Maybe it had some effect in 2016 but that effect might have been relatively small and varied from state to state. This explanation can be immediately dismissed with more granular data. I mentioned Connecticut, which has a high education rate despite Clinton’s comparatively poor showing there. Township-level data suggests that there was an immense education divide, with Clinton improving as much as 40% on Obama’s performance in affluent, highly-educated suburbia while declining by as much as 30% in rural areas with low rates of college education. Connecticut is a good example of this phenomenon but it is by no means the only one; we see this pattern emerge in the localised data of every single state in America. So if the education gap is real, why didn’t it show up on the state level? Could it be because of another demographic feature, such as the urban/rural divide? Or ideology perhaps? Neither of these seem to work either. Clinton improved with whites in liberal states (CA) and conservative states (LA), in urban states (WA) and in rural states (KS). Similarly, Trump improved in liberal states (RI) and conservative states (WV), in urban states (NY) and in rural states (ND).

The answer is somewhat simpler, if less immediately apparent. The play for white voters in 2016 was a tug-of-war, of sorts. Trump was pulling non-college-educated whites to the right, Clinton pulling college-educated whites to the left. The trouble for Trump was that, in some parts of the country, no matter how hard he pulled, he couldn’t gain much support with non-college whites, the very demographic at the core of his campaign. This is because gains in voting margin are inherently logarithmic, not linear, although it can be easy to conflate the two. Getting from 80% to 90% with a certain group is much easier than getting from 90% to 100%, although the gains are equal in absolute terms. The former requires winning over half of voters who aren’t already with you, the latter requires winning all of them. Why was this relevant in 2016? Well because in states like Louisiana and Mississippi, Mitt Romney had already won nearly 90% of non-college-educated white voters back in 2012. There was little room for Trump to grow with this demographic, no matter what he did. Don’t get me wrong, Trump did make some gains with southern non-college-educated whites in places like the Louisiana Bayou, North-East Mississippi and Northern Alabama. But the effects of those gains were muted in comparison to states like North Dakota and West Virginia where the President started from a much lower baseline.

At the same time, college-educated white voters across the country swung left, regardless of geography or ideology. In a state like Louisiana, even though college-education levels are relatively low, Clinton had ample opportunity to improve with this demographic from Obama’s incredibly low baseline of ~15%. In the end, she achieved exactly that and ended up winning a higher share of the Louisiana white vote than Obama did, despite Louisiana being a heavily non-college-educated conservative state. That, right there, is the power of brick walls. All the targeted campaigning in the world can’t help you escape the fact that once you hit 90% with a certain demographic, you’re basically stuck.

In 2016, this whole dynamic actually served to Trump’s advantage, exacerbating the electoral college/popular vote divide. It didn’t really matter if he slid a couple points with Southern white voters; they’re not the ones who cast the tipping-point votes. That’s the Midwest. In 2020 though, I wouldn’t be so sure. Whatever troubles Trump had with Southern whites in 2016 look to be a lot worse in 2020, and uniquely concentrated in the recently-minted swing states of Texas and Georgia.

Let’s take Georgia, for instance. Looking only at the two-party vote, I estimate that Romney won 79% of college-educated whites, whereas Trump won just 64%. A defection rate of 19%. Similarly, Obama won ~18.5% of non-college-educated whites, while Clinton won just 13%. A defection rate of 30%. As you can see, the defection rate was much higher for non-college-educated white Democrats than for college-educated white Republicans but the relative size of the groups ultimately meant that Clinton did 5-6 pts better than Obama with white voters in Georgia. Combined with a drop in African-American turnout, that was enough for her to cut Trump’s margin in Georgia to just 5%. Looking pretty purple.

Without taking demographic shifts into account, the picture looks pretty grim for Republicans in 2020 if the same trends of defection by education occur as in 2016. Such a trend would cut Trump’s statewide margin to just 1.3%, to the point where a slightly more diverse electorate could tip the balance. A reminder that this is also a scenario where Democrats win the popular vote by quite a narrow margin nationally.

At this point I’ve conclusively shown that a growing education gap is damaging to Republicans in the South. They have a huge downside with college-educated voters and a minimal upside with non-college-educated voters. This issue for the GOP could in part be remedied by higher turnout levels, as non-voters skew less educated. However they also skew less white so it’s not clear that would be the boost that Republicans want. So now I guess the pertinent question becomes; will there be an increase in the education gap, specifically among white voters, in 2020? I believe the answer is yes, and I have a couple of ways of thinking about it.

The first gets to heart of what Trump is about. It’s clear that a major reason for the explosion in the education gap in 2016 is Trump’s persona, his bombast and his bigotry. Nearly everything he said or did served to alienate college-educated white Republicans, meanwhile their non-college-educated counterparts couldn’t get enough of it. It’s also pretty clear that since November 2016, Trump has doubled down on ‘Trumpism’. Whether it’s Kavanaugh, impeachment or locking children in cages, Trump has only ever thought of political repercussions in terms of his base, and he hasn’t shied away from embracing everything that made him controversial back in 2016. So if we can draw a link of causality from Trumpism to the education gap in 2016, I don’t think it’s a stretch to say we should be able to draw a similar link from heightened Trumpism in the intervening years to an even larger education gap in 2020.

The second is purely data-oriented and it seems to support the first. Across a variety of polls and elections, there is a body of evidence to suggest that when it comes to ‘culture war’ issues – the kind of things Trump will surely bring the forefront of his 2020 campaign in his efforts to hold onto WI/MI/PA – it’s college-educated whites who are the first to abandon the GOP. A sampling:

  • In 2017, the Alabama Republican Party nominated a racist pedophile who went on to lose what should’ve been a shoe-in Senate race. The Democrat, Doug Jones, improved on Hillary Clinton’s 2016 performance with every demographic group but made his biggest gains by far with college-educated whites. Similarly, a number of Republican voters couldn’t stomach the idea of voting for Moore, the pedophile, but didn’t want to vote Democrat and so cast a write-in ballot. These write-in votes were predominantly cast by college-educated whites in places like the suburbs of Birmingham and Huntsville.
  • The 2018 election that most resembled what the 2020 Presidential election will probably look like was the race for Senate in Texas between Beto O’Rourke and Ted Cruz. It featured incredible levels of ad spending, a heavy focus on ‘culture war’ issues that have defined the Trump era, and an unexpectedly high turnout. Once again, O’Rourke ran well ahead of Clinton with educated white suburbanites, while staying even or even falling further behind with rural non-college-educated whites. One highly educated suburban county of Dallas, for example, when from Romney +31 to Trump +17 to just Cruz +6.
  • Broadly, across most House races in the country, Democratic candidates in 2018 did better in heavily educated areas than in 2016. This was true of many races that weren’t even remotely competitive.
  • When we use special congressional elections to look at trends beyond the normal two-year cycle, we can see that the Democratic advantage with educated voters has tended to grow over time. Jon Ossoff famously failed to flip the heavily educated GA-06 in the April 2017 special election, but Lucy McBath won it in November 2018. When Dan McCready ran again for NC-09 in a special election in September 2019 after the midterm election result was declared invalid, he did worse in every single county except for Mecklenburg, which was easily the county with the highest percentage of college-educated whites.
  • In all three of the Gubernatorial elections in 2019 but particularly in Mississippi and Louisiana, Democratic candidates for Governor pulled off record margins in affluent/educated suburbia in competitive races. The fact that Jim Hood, a Democrat from rural Northern Mississippi was able to win Madison County (an educated suburb of Jackson, MS) despite lethargic black turnout is a testament to how much of a drag Trump/impeachment is on Republican candidates in these areas. I can tell you, Hood does not have a unique personal appeal to suburban Republicans. He didn’t even win Madison County in his 2015 landslide win for Attorney General! What changed? Trump. A similar story can be seen in New Orleans, where incumbent Democratic Governor John Bel Edwards won an impressively large number of educated suburban Republicans. Prior to the election, I asked Louisiana pollster and elections analyst John Couvillion if he thought Edwards could improve on his 2015 performance in Orleans Parish, by far the most college-educated parish in the state (among whites). He said no; I disagreed. In the end, Edwards got 90% of the vote there, up three points from his 2015 numbers. In doing so, he won college-educated white voters in Orleans by such a wide margin that it would’ve been largely unthinkable four years prior.
  • An analysis of NYT/Siena polls from late 2018 show that the education gap on ‘culture war’ issues such as the NFL protests, reverse racism and gun control is nearly uniformly wider than the education gap on Trump’s approval rating. In other words, there are a fair few anti-Trump voters who side with him on cultural issues and a number of pro-Trump voters who don’t side with him on cultural issues. The former tend to be non-college whites, the latter tend to be college-educated. If ideological sorting by these issues continues as we saw in 2016, we can expect a wider education gap in 2020.

On the face of it, whether or not the 2020 election features a wider education gap than in 2016 might seem to be a meaningless trivia question. And if you think the election revolves around Wisconsin, as so many do, it is exactly that. Who cares about what coalition Trump has, all we care about is the top line result. We care who wins. But as I hope readers will now see, the expansion of the education gap could/will have a profound effect on who the next President will be. If the Trump playbook works well for the President in the Rust Belt, it could prove to be his Achilles’ Heel in the Sun Belt, if Democrats are willing to exploit it. Of course, none of these trends are set in stone, it is far too early to make concrete predictions about what will happen in November. But there are strong reasons to think that Trump will fall further with college-educated whites and strong reasons to think that such a shift could be his undoing in the South. Don’t be surprised if you wake up on November 4 to find that Trump successfully defended his trio of Rust Belt states only to lose re-election because voters in the Lone Star State who have four-year college degrees have finally had enough of his shit.

Don’t Be Fooled By The Polling Averages – Sanders Leads In Iowa

If you’re a serious elections junkie you probably follow the work of Nate Silver and his colleagues at FiveThirtyEight. You might even check their primary forecast every few hours to see how the odds have or haven’t changed, as I did in the lead up to the 2016 general election. If you are that person, firstly may I say that you’ve picked something very cool and important to obsess over, so good job. Secondly, maybe it’s worth taking a bit of a break from the impressive but ultimately flawed work of polling aggregators like FiveThirtyEight and RealClearPolitics. Their outputs say that Joe Biden and Bernie Sanders are tied in Iowa. It’s not the most incorrect proposition ever, but it is needlessly under-confident. Sanders leads in Iowa. Here’s why the averages are misleading:

1. The variation in Iowa polls is not due to simple randomness but rather distinct methodological differences

When Monmouth University (A+ rated by FiveThirtyEight) finds Biden up six points in Iowa in the same two-week period that Selzer & Co (A+) find Sanders up three (with Biden in fourth place) and NYT/Siena (A+) find Sanders up seven, that’s not just due to sampling variation. What’s happening here, as expertly explained by NYT’s Nate Cohn is that Monmouth’s likely voter screen is excluding voters who say they’re likely to caucus but didn’t vote in recent primary elections or the 2018 general election. Selzer and Siena, on the other hand, rely more heavily on voters’ self-reported intent to vote. To illustrate this point, Cohn recently showed that under Monmouth’s methodology, Siena would’ve found Biden leading Sanders 24-23. One in five voters in the Siena poll did not pass Monmouth’s screen, and these voters backed Sanders over Biden 42-10. Sanders will win Iowa if a) these voters turn out as Siena/NYT expects them to and b) there is no late break against him. B) is an open question that we can’t predict very well and it’s why I can’t say for sure that Sanders will win, merely that he is the favorite to do so. But a) is a question we can meaningfully answer and should answer when it comes to predicting the winner. To Cohn’s credit, he does not say his findings are rock solid, he accepts with humility that pollsters such as Patrick Murray of Monmouth may be the ones in the right and he may be eating humble pie come February 3. I would go further, however. Cohn’s methodology IS the right path to take, regardless of whether Bernie actually ends up winning Iowa. Excluding (overwhelmingly young) voters who say they’re likely to caucus simply because they haven’t voted in previous primary or midterm elections – which are quite different to caucuses – is not good polling practice. That isn’t to say Monmouth isn’t a good pollster, they are and they have done amazing work in the last few years. But in this particular instance I am firmly on the side of Siena/NYT and therefore the only reasonable conclusion I can make is that Bernie Sanders has a robust lead over the field in Iowa.

2. Sanders has a strong positive trend

Whenever a new poll comes out in Iowa these days, like clockwork, Sanders is doing better than previously. I mentioned before that Monmouth University had Biden up six over Sanders in their early January poll – yesterday that lead was cut to two points. A recent Iowa State University poll showed Sanders at 24%, up from 21% a month ago. Suffolk University had him at 19%, up from 9% (!) in October. And Siena/NYT had him at 25%, up from 19% in November. It’s always good to head into the final week of an election with a positive trend line but it arguably matters even more in Iowa, where voters can often change their minds at the last minute, giving an unexpected boost to the candidate with the most ‘momentum’. In 2016, it was Marco Rubio who was on the up-and-up in Iowa polls – he was at 11% for most of December/January but surged to 16% in the final week. That momentum continued to caucus night and he ended up finishing a surprisingly close third with 23% of the vote. Or take the classic example of Rick Santorum in 2012; he burst from sixth place in the two weeks before the caucuses and increased his vote share in polls from just 6% to 16%. In the end he actually won Iowa – although by just 0.1% – despite polling six points behind eventually nominee Mitt Romney. This matters now because it seems Sanders is the only leading candidate with a clear positive trend, although perhaps not quite as strong as that of Santorum or Rubio. Biden, Buttigieg and Warren – all of whom could feasibly win the caucuses – are all flat or trending down. The only other candidate who could claim undeniable upwards momentum at this stage is Amy Klobuchar, but her gain is also Sanders’ gain, who would love it if she took more votes from Biden/Buttigieg. More on this in a bit.

So while Sanders’ raw position in the polls is good but not incredible, his trend line is something his campaign should be very happy with. Being in first place AND moving upwards with five days to go is a damn good combo.

3. Precinct thresholds may help Sanders (or: the Klobuchar singularity approaches)

The recent polls with the best data on this questions (Siena & Monmouth) are both in agreement; Sanders does better against Biden once candidates outside the top five (Biden/Sanders/Klobuchar/Warren/Buttigieg) have their choices re-allocated. Monmouth has the race going from Biden +2 to a TIE in a six-way race (the five + Yang) and would presumably show a Sanders lead of ~2 if Yang’s choices were re-allocated too. Meanwhile Siena has the race at Sanders +7 both on the ballot with all the candidates and the four-way ballot (five minus Klobuchar). If Klobuchar were still in the running, it seems likely Sanders would extend his lead out to ten points in the Siena poll.

This matters in Iowa in particular because each precinct has a viability threshold of at least 15% so if your candidate cannot reach that threshold, they get no votes (a.k.a State Delegate Equivalents) from that precinct. As candidates such as Yang and Steyer will likely reach 15% in very few precincts, if any, the second choice intentions of their supporters are important. While Sanders seems to benefit from picking up Steyer/Yang/Gabbard supporters, his major weak point is Klobuchar. By all accounts, he’d be lucky to pick up 10% of Klobuchar voters in precincts where she’s not viable, whereas Biden might be earning a majority of those voters. This means Sanders has got to hope Klobuchar is viable in as many precincts as possible, to cut into Biden’s support. The good news for the Sanders camp is that this seems to be materializing. Recent polls show Klobuchar gaining statewide, garnering around 10% of the vote. And new crosstabs from none other than Nate Cohn confirm what seems intuitive – Klobuchar’s voters are disproportionately clustered in certain precincts such that she will reach 15% in many of them even if she falls short statewide. His Siena/NYT poll had Klobuchar doing nearly twice as well in Clinton ’16 precincts (11%) than Sanders ’16 precincts (6%). If I had to guess I’d say Klobuchar ends up being viable in 20-40% of precincts – not exactly the late surge the Sanders campaign would love but enough that she can be a thorn in Biden’s side.

At the same time, Warren being non-viable in a lot of precincts could also give the Sanders campaign a boost – not that it necessarily needs one. Recent polls show her hovering at around 15% statewide, enough to be viable in most precincts once undecided voters have been allocated. However it still seems likely she will miss out on viability in a number of precincts – particularly rural or conservative ones. Just as Klobuchar is on the cusp of viability in precincts that voted for Clinton in the 2016 caucus, Warren is in the danger zone of non-viability in these precincts, getting just 12% there. This too helps Sanders because he is the clear second choice of Warren voters, although he likely won’t earn a majority of their votes. Warren being non-viable in 20-40% of precincts could give Sanders a boost of maybe one percentage point, and that’s nothing to scoff at in a close race.

Finally, Sanders benefits from having quite well distributed support around the state such that he is unlikely to fail to reach the 15% threshold in many precincts. In the Siena poll he is at 31% in Sanders ’16 precincts and at 23% in Clinton ’16 precincts. He gets at least 19% in all four ‘regions’ of Iowa. Admittedly he only gets 9% of the 65+ voter demographic but luckily that group doesn’t cluster in the same way youth do around universities. There won’t be many precincts where 65+ voters are a majority. Biden on the other hand, only cracks 15% with 65+ voters and is at just 10% with 18-29 year olds. Those numbers lead me to believe he will almost certainly be non-viable in a few precincts in liberal strongholds where Sanders did well in 2016 such as Iowa City and Fairfield.

Sanders is the likeliest candidate to winIowa no matter what weird stuff happens with precinct viability, owing to his current polling lead, enthusiastic base and upwards momentum. Even in a world where Klobuchar is completely non-viable and Biden/Warren/Buttigieg are completely viable, he maintains a 7-pt lead per Siena. But it seems to me that the current dynamic lends him even more strength, as the moderates candidates split the field in their scramble to stop him, blinded by their own desire to be President and inability to understand just what draws voters to Bernie Sanders. Klobuchar shares a lot more in common ideologically and strategy-wise with Biden than with Sanders but she may just end up handing Iowa to Bernie just s Michael Bloomberg might hand a state like Texas to the Vermont Senator on Super Tuesday. Everyone in the race really truly believes they should be President and for that reason there isn’t really going to be a concerted effort to stop Sanders. If anything, Biden losses in Iowa and New Hampshire might cement the idea that he can’t stop Sanders, pushing even more voters towards a moderate alternative (Bloomberg, Klobuchar, Buttigieg, Bennet? who the fuck knows, let’s wait for the next Jonathan Chait op-ed to inform us). Ironically, this would probably only bolster Sanders’ campaign as we head into the delegate-rich primaries of March and April. But I’m getting ahead of myself. The point is; the conditions in Iowa are ripe for a Bernie Sanders victory. Don’t kid yourself into thinking it’s a tossup just cause Nate Silver says so.