Tuesday 19 March 2024
.
This week saw a significant milestone in the 2024 elections, as Joe Biden and Donald Trump were mathematically confirmed to be their respective parties' nominees for the Presidential race, making it official that 2024 will see a rematch of the 2020 race. With the stage set, now seems a good time to take stock and provide a way-too-early preview of where things stand, and what we can expect as we head into the campaign.
Setting the scene
While this week's primaries maybe have technically confirmed each party's presumptive nominee, in truth there has been little question of the outcome for some time. As the incumbent President, there was never any serious doubt that Biden would be nominated by the Democrats and, similarly, Trump's stranglehold on the Republican Party is such that his nomination, too, has seemed a foregone conclusion - although the level of anti-Trump sentiment during the primary has been surprising, more on that later.
While this may have seemed a foregone conclusion to political analysts, polling has consistently shown that most voters did not believe, or realise, that these two would end up being the candidates. This is a fascinating little quirk that illustrates the stark disconnect between political coverage and the average voter. While the politically engaged like to obsess over every slight news event and extrapolate how every headline might affect the electoral outcome, the truth is that most Americans are so disengaged that they weren't even aware of the overwhelming likelihood of this matchup. I flag this, as it will become relevant when we consider the predictive value of current polling.
One way or another, barring some unforeseen event, this is the Presidential race we have. So who will prevail? That is what I intend to consider with the rest of this article. It is, of course, far too early to make any kind of reasonable predictions or forecast. That will come closer to election day. But we can at least provide some general sense of the direction in which things are heading, and how they are likely to change in the coming months.
It is also worth mentioning that, in addition to the Presidential election, 2024 will also see contests for the House and Senate. For the purposes of this article, I will be focusing on the Presidential election specifically, although I will provide a brief summary on the state or play for these legislative elections.
What do the fundamentals say?
In analysing any election, the logical place to start is the beginning. What is our starting point heading into this election? What priors and presumptions can we draw from history, an understanding of electoral mechanics, and the general macroscopic state of the nation, before even looking at the current data?
The first thing to note is that incumbency advantage is a very real thing in Presidential elections. Incumbent Presidents usually win. In fact, in the last 100 years, only 5 of 19 Presidents have failed to win re-election, including one of this year's candidates, Donald Trump. This is a well-studied phenomenon. Partly it is because incumbents will hold greater name recognition, but it also owes to the fact that voter preferences tend to be surprisingly fixed. Voters don't like to second-guess themselves or admit a mistake. So if they voted for someone previously, they are highly likely to do so again.
This effect is even more pronounced in political rematches. We don't have many examples of this in Presidential races, but in Congressional races it happens much more frequently than you would think, and almost always ends the same way as the previous contest. The most recent example we have of this at the Presidential level would be the 1956 election, which also ended the same way as the preceding contest. Again, this likely is due to voters becoming entrenched in their opinions, but there's also a certain taint that comes with being an electoral loser. Voters like a winner. This is the reason why Trump spends so much time talking about winning, and why he has consistently pretended to have won the 2020 election, which he didn't.
It is also worth highlighting that this being a Presidential election year likely helps Democrats. In recent history, Democrats have tended to perform better in Presidential election years than in midterms (see later - some suggestion that this may be changing). These elections tend to have higher turnout among young and more demographically diverse voters, who tend to vote heavily Democratic.
What can we glean from recent election cycles? This has been surprisingly under-reported, but the Trump era has largely been a period of decisive electoral victories for Democrats.
Democrats won the 2018 midterms by historic landslide margins, no mean feat considering they have historically performed so poorly in midterms.
In 2020, Biden defeated Trump by a wide electoral and popular vote margin, expanding the map into states like Arizona and Georgia that Democrats haven't won in a generation. That they did this against an incumbent President (incumbents, as noted above, rarely lose) is a remarkable overperformance that for some reason hasn't really been talked about.
Then we have the 2022 midterms. The President's party usually suffers heavy losses in a midterm, especially if they're a Democrat (due the demographic disadvantage, noted above). Yet somehow, Democrats actually made gains in the Senate, and came very close to holding their House majority despite predictions of a red tsunami.
All three of these election cycles represent historic over-performance from the Democrats, by any reasonable metric. More notably, they arguably demonstrate diminishing electoral returns for Trumpist Republicans with each cycle.
Then there's the economy. Unemployment is consistently at record lows. While inflation and gas prices have been high in the post-pandemic recovery, both have been consistently declining under Biden. By any metric the economy is booming right now, a very strong indicator for Biden.
A common refrain in a Presidential election is to ask voters if they are better off than they were four years ago. In this respect, Biden has arguably the easiest pitch in the history of Presidential politics. Four years ago, America was in the midst of a pandemic that killed millions, with an economy in freefall and double digit unemployment. Trump left office with an economy in ruins and the ignominy of being one of the only Presidents in modern history to see a net decline in jobs during his tenure. By any metric, the country is in a far better shape now than it was four years ago
Lastly, let's consider Biden's approval ratings. These are, it has to be said, fairly mediocre. Most approval ratings released over the past month have him in the low 40s (the average of the 9 most recent polls on 538 comes to 42%, with a range from 39% to 47%), which is historically pretty middling. It's higher than Trump, who spent most of his Presidency with approval in the 30s, and lost, but lower than Obama, who did have low 40s approval ratings in his election year, but started to push into the high 40s by this point and remained there until election day, which he ultimately won fairly comfortably. Approval ratings are far from an exact science, and not massively predictive, but one would typically expect a candidate with approval lower than the mid 40s to struggle. It is not clear how well that holds when up against a competitor with even worse approval ratings. It is also worth noting that incumbent approval ratings tend to improve as we get into election year, so watch this space.
In summary, we have an incumbent, overseeing a booming economy that has recovered significantly from one of the darkest eras in American history, running in an election cycle that history tells us will be demographically favourable for his party, on the back of a historic electoral overperformance in a cycle that was demographically less favourable than the current one. Approval ratings are a mild concern, but we would expect them to improve, and they aren't a mile off from where we would expect a winning incumbent to be.
This establishes a baseline expectation that 2024 "should" be a good year for Democrats. Only, this hasn't been the story of the 2024 election so far. In fact, most coverage has been painting the exact opposite picture, one in which Trump has the advantage. Why is this the case, and is it a valid depiction of the current state of the race?
What do the polls and other data say?
In short, the current Biden pessimism is coming from the polls. You will no doubt have seen the media breathlessly pointing to the polling aggregators. RealClearPolitics currently has Trump up 1.6%, DecisionDeskHQ has him up 1.2%, Race to the WH has him up 1.4%. Other aggregators, in fairness, are not showing this - The Economist and PollingUSA both currently have Biden ahead on aggregate. But the question here is about perception, and for whatever reason it is typically the aggregates that show Trump ahead which get the most media coverage.
In short, the current Biden pessimism is coming from the polls. You will no doubt have seen the media breathlessly pointing to the polling aggregators. RealClearPolitics currently has Trump up 1.6%, DecisionDeskHQ has him up 1.2%, Race to the WH has him up 1.4%. Other aggregators, in fairness, are not showing this - The Economist and PollingUSA both currently have Biden ahead on aggregate. But the question here is about perception, and for whatever reason it is typically the aggregates that show Trump ahead which get the most media coverage.
I have a few things to say about this, as the media is generally quite poor at reading and reporting on polls, and right now they are ascribing far greater value to this metric than they should be.
But before we dig into the specifics of the polls, I want to look at the rest of the available data. Pundits often treat the polls as the one and only source of data when discussing electoral prospects, but this is not the case and there are a number of other metrics that have predictive value.
First up, Congressional retirements. Elected officials from one party retiring or declining to run for re-election in large numbers is generally a bearish sign. It indicates that those officials have low expectations for their electoral prospects or the state of the party as a whole. Right now, the Republican party is seeing a mass efflux of Congressional members, which is not being matched by the Democrats.
Fundraising depicts a similar dynamic, where Democrats and Biden are significantly outraising Republicans and Trump, and indeed there are numerous stories of state Republican parties being on the verge of insolvency in key states like Michigan.
But probably the most important additional data point I want to consider, one which is often overlooked in media, is the actual election results. With all the talk of polls, it's easy to forget that there are actual special elections taking place all throughout the year, some as recently as just a few weeks ago.
Once again, Democrats have been performing strongly here. Democrats have been outperforming in special elections relative to their baseline - ie, relative to what one would expect in a break-even year based on the partisanship of the electorate. In fact, currently they are outperforming their baseline by 11% on average. This is a very important point to note. Special election overperformance was one of the key indicators in 2022 that Democrats were headed for a good year, one which most pundits ignored.
It is also important to note that Democrats have not just been outperforming their baseline, but also the polls. Taking the most recent special election, NY-3, as an example, the polls had the race at D+3 on average. The actual result was D+8. We can see this consistently across special elections. Democrats are not just performing better than we would expect from their baseline, but also significantly and consistently outperforming their polls.
Lastly, let's briefly consider the primary elections. I am reluctant to read too much into primary election results, as they generally are not predictive of the eventual general election. I don't, for example, think you can make any conclusions by the number of people who voted in the Democratic primary vs the Republican primary, or the margins in general. What is interesting, however, is to look for consistent patterns in polling errors, and we have seen exactly that in this year's primaries.
Polling has significantly and consistently underestimated Biden's vote share and overestimated Trump's vote share in these primaries. Most polling was showing Biden with 60-70% in the primaries, a lacklustre result for an incumbent. He generally ended up with 80-90%+. It was a similar story on the Republican side, where Trump was typically polling 80-90%, but mostly ended up with 60-70% in most races, and even ended up losing a few contests to Nikki Haley. If you consider that Trump is running as, essentially, an incumbent, then those results could be a bit of an alarm bell for November. The primary vote data actually gets worse for Trump the more you look at it, with particular weakness in the crucial suburbs that swung the 2020 election, and which Trump desperately needs to win back in 2024.
So overall we have some fairly mixed data. Fundamentals paint a positive picture for Biden, as does most of the data (retirements, fundraising, election results), but then the polling right now ostensibly is showing Trump in the lead. How do we square this apparent discrepancy? Let's have a closer look.
Analysing the data
The first thing to note, with respect to the polling averages, is that a 1.5% margin is not statistically significant. Most polls have a minimum margin of error of 3-4%. So even if you accept that the polls have Trump ahead by around 1.5%, this falls well within that margin. In other words, what these polls are really saying is that there is 95% confidence that the current position lies somewhere within Trump +5 to Biden +4. Not exactly definitive.
The second, and most important thing to note is that we are still early in the election cycle, and polls at this stage are generally not very predictive - in fact, they are less predictive at this stage than some of the other data referenced above, such as special election results.
This shouldn't come as a surprise. As noted earlier, most voters aren't even aware that these two are the candidates at this point. They just aren't paying attention yet, much less in a position to have established a firm intention as to November. More to the point, it's not surprising to see the challenger leading at this point in an election cycle. Gallup, ABC, Washington Post all had Romney leading at this point in 2012, for example. This makes sense, as the challenger will have already been well into their campaign since the start of the primary - in this case, Trump has been campaigning since the end of the last election. Biden, by comparison, has been busy running the country, and is only just now beginning to switch into campaign mode.
In fact, about the most consistent pattern we can see historically in relation to early polls is that the incumbent tends to improve their numbers in the months leading up to the election, likely for the reason noted above. So, even if you believe that Trump is about 1.5% ahead in the current polls, history tells us that the numbers are likely to improve for Biden as we get closer to the election. Given the very narrow margin, that could easily swing the advantage towards Biden.
It is also worth noting that most polls right now are using what they refer to as a registered voter model, rather than a likely voter model. This means that the polls' current turnout model may not accurately represent the likely electorate in November. One of the commonly proposed explanations for why the polls don't reflect the actual election results is the idea that Trump is performing well with low-propensity voters who are less likely to vote, the idea being that they didn't turn out for special elections, but might in the general election. If this is correct, then you would expect the likely voter models will remove a lot of these less-likely voters, and move the numbers more in Biden's favour. A recent example of this in practice is this week's new poll from Florida Atlantic. Last month they had Trump up by 4%, this month they switched to a likely voter model and Biden is up by 2%. As you can see, the potential swing from this switch is very significant.
So there is good reason to think that poll numbers are likely to improve for Biden in the coming months. But I'll go further, there is also very good reason to think that the current polling averages may not be an accurate depiction of the current state of the race.
To demonstrate this point, let's revisit the 2022 election. The 2022 election is generally considered to have been a massive shock, one of the biggest polling misses in recent history. Polling averages, and the pundits that read them, were emphatic that Democrats would be decimated. When that did not happen, everyone was shocked. Only, they shouldn't have been shocked. I wasn't shocked. In fact, the uncannily accurate preview on this very blog predicted a close contest in the House and Democratic gains in the Senate. This wasn't just wishful thinking, it was based on the data. Data that most in the media either ignored or misinterpreted.
What happened in 2022 was that there was a material disparity between the established, reputable, high quality pollsters (Quinnipiac, Ipsos, Monmouth, Morning Consult, The Economist, Marist), who were generally showing more positive numbers for Democrats, and the other lower quality pollsters (Rasmussen, Trafalgar, Harris), which generally showed the massive Republican gains that never materialised. This was the story of the 2022 election. High quality polls did, in fact, predict the strong Democratic performance, but they were drowned out in the averages by the much larger volume of, frankly, low-value data.
It seems very few in the media actually learned anything from 2022, and the narrative of that election cycle has largely been dumbed down to "the polls were wrong". The idea that, in actual fact, some polls were pretty accurate in 2022 requires more nuanced analysis and critical evaluation than they are willing, or able, to provide.
This is crucial to our understanding of the 2024 election, because it seems that very much the same thing is happening. The Economist recently did an analysis showing that, once again, there was a material disparity between the high quality polls, which were more favourable for Biden, and the low quality ones, which were more favourable for Trump.
In fact, if you look at the recent polls, it's really only the likes of HarrisX and Rasmussen, pollsters with a very long and consistent track record of having a "house effect" (read: bias) towards Republicans, propping up this Trump advantage. Most of the recent polls from high quality pollsters (Ipsos, Reuters, Quinnipiac, I&I/TIPP, Civiqs, Emerson, Morning Consult) all show Biden in the lead. Of the high quality pollsters, it's really only the one NYTimes poll showing a Trump advantage at the moment.
One last observation on the current polls. In general, I am hesitant to dig too much into the underlying data in the crosstabs. I've seen a lot of people in the media digging into these numbers and drawing all kinds of wild conclusions (on both sides) - more on this later. But there is some useful data to be gained here, specifically where they have polled individual policy issues. In particular, I note a lot of these more Trump-favourable polls are also finding significant majority support for things like nationwide abortion bans and shutting down planned parenthood, things which have never been popular in the past. If you show me a poll that says a particular candidate is more or less popular than before, I may not buy it, but it is plausible. The idea that a majority of the country could suddenly completely change their values and beliefs on historically entrenched topics like abortion, is simply not plausible to me. This, to me, is a red flag that those polls may be over-representing Trump supporters, who already hold those beliefs.
To be clear, this is not about predicting polling error based on the last election, which is notoriously unpredictable and changes with each election cycle. This is simply to point out that some pollsters are more reliable than others, and currently much of the Trump-favourable poll numbers are coming from pollsters with a poor track record.
So it seems quite plausible that this current Trump advantage in the polls (slight as it is) may simply be another 2022-style mirage. Accordingly, this apparent disparity between the polls and other data may end up not being such a mystery after all, and just vanish as time goes on.
Conclusions
Going back to the question posed at the end of this piece's introduction: is the current media depiction of this race as one where Trump is winning fair and accurate? Probably not.
Going back to the question posed at the end of this piece's introduction: is the current media depiction of this race as one where Trump is winning fair and accurate? Probably not.
This narrative appears to be based on a fairly superficial and un-nuanced interpretation of just one out of several data points, and completely ignores other valid data.
Taking a more holistic view of the available data, these apparent inconsistencies make a lot more sense. The fundamentals and most data are perfectly consistent with an election year that favours the Democrats. The polls, when subjected to a bit of critical scrutiny, also would seem suggestive of a more favourable environment than the aggregated average currently shows.
Make no mistake, this is a close race, even if it is likely leaning towards Biden at the moment. A lot can change in the next eight months. But unless something major happens, like a health scare or scandal, my current expectation is that Joe Biden remains most likely to win the election.
This goes both ways, though. Trump being the same age as Biden, could equally have his own health scare. We also haven't considered the giant elephant in the room of Trump's ongoing criminal prosecutions. It's entirely possible that Trump could be a convicted felon by the time of the election, and even if he isn't, his trials will be occupying a huge amount of media coverage, not to mention Trump's time that he could otherwise be spending on the campaign trail. So while it is fair to say that Biden has potential downside that could turn this election against him. You would have to say that the potential downside is far greater for Trump.
To me, the current state of the race is clear. You would expect Biden to have an advantage over Trump, and the present data seems to support that this is the case. I don't think you can reasonably argue the alternative at the moment, without cherry-picking and ignoring valid data. It also seems most likely, based on historical patterns and the potential exposures both candidates face, that the race will continue to move towards Biden over the coming months. In short: you'd rather be Biden than Trump right now.
Debunking a few myths
I wanted to dedicate a segment of this article to debunking a few common myths, prevalent in how those in the media and social media tend to analyse election data.
I wanted to dedicate a segment of this article to debunking a few common myths, prevalent in how those in the media and social media tend to analyse election data.
Myth: the polls show Trump support surging among young voters/minorities/[insert demographic here].
Answer: as noted above, there's been a lot of coverage in the media about the underlying data in the polls. NYTimes recently ran an article talking about Trump's historic surge in support among young voters. Others have observed Trump making historic gains with black and hispanic voters, a result that, if true, would represent the most significant racial realignment since the civil rights era. The problem is, it probably isn't real.
The cross-tabs in polls are notoriously messy, and commonly show wonky results. This doesn't mean the top-line result of those polls is not accurate, often they are even when the cross-tabs look strange.
More often than not, reading too much into this data will result in embarrassment. With the NYTimes piece, for example, NYTimes ran this huge, headline press-release about the youth-vote finding from their first poll, only for their follow up poll to show the exact opposite result (a finding which, strangely, did not merit its own headline). Regarding these claims of a historic racial realignment, even aside from the fact that this seems unlikely on the surface, it's also worth noting that polling in recent election cycles has shown similar shifts, that ultimately did not materialise in the general election. In 2022, for example, Democrats actually performed better with hispanic voters than in 2020, despite polls showing a big shift towards Trump.
At the end of the day, big claims require big evidence. Better evidence than notoriously erratic cross-tabs. We see this kind of crazy cross-tab shifting every election. It usually amounts to nothing. Until we get some actual data verifying these claims, let's maintain a healthy skepticism. It's a classic example of not worrying about how the sausage gets made.
Myth: the polls always underestimate Trump, so we should assume he will outperform the polls in 2024.
Answer: I've seen a lot of this kind of suggestion, including from people who should know better. The idea is that Trump outperformed his polls in 2016 and 2020, and so we should assume he will do so again this year. As such, if he is roughly tied in the polls or if Biden leads narrowly, we should assume Trump is actually winning.
This is, of course, nonsense. Despite what you may have read, the polling errors in 2016 and 2020 were not actually all that unprecedented, nor were they all in the same direction in every race. For example, Trump outperformed 2020 polls nationally, but underperformed in certain key states like Georgia and Arizona. Trump-affiliated politicians also significantly underperformed polls in both 2018 and 2022, and Trump himself has significantly underperformed his polls so far in 2024.
Bottom line, polling errors are notoriously difficult to predict and usually change from one election cycle to the next, even with the same candidates. Polling error in one election does not suggest there will be a similar error in the next. In fact, usually the opposite is true as pollsters overcompensate for their previous bias. There is no particular reason to believe that Trump will outperform his polls in 2024, and in fact some evidence to suggest the opposite.
Myth: Biden needs to win the popular vote by 4%+ due to the electoral college.
Answer: this myth is based on the observation that Hillary Clinton won the national popular vote in 2016, but lost the electoral college, and Biden won the national popular vote in 2020, but won key states only narrowly.
This is not a particularly strong analysis. It is well observed that the electoral college advantage changes with each election cycle. It is not especially predictable, nor interconnected enough that you can say an X% decrease nationally correlates to a Y% decrease in a particular key state. While it is true that Biden likely would have lost in 2020 had the national popular vote been a tie, the opposite was true in 2012 when Obama had this advantage over Romney. Similarly, in the most recent election cycle in 2022, Democrats won statewide races in key states like Pennsylvania, Michigan, Georgia and Arizona despite losing the national popular vote. So it can and will change, and we can't assume Biden needs to win the national popular vote by any particular margin. In fact, there is evidence to suggest that the electoral college advantage will be narrower this year, and that Biden may only need a small national popular vote advantage, if any.
Myth: The discrepancy between the November polls and recent election results is explainable by Trump's strength with low-propensity voters who are more likely to vote in November
Answer: I've already addressed this particular line of thinking earlier in this piece. The idea here is that the polls are so much more pro-Trump than the actual election results so far because Trump is polling disportionately well with voters who are less likely to vote in special elections and primaries.
I want to be careful here, as this isn't a terrible theory and there's no hard proof against it. It is, after all, a theory based on the absence of data, so there is fundamentally no way to prove or disprove it. There are, however, a few things about this that don't hold water. It does nothing to explain why the polls for the primaries and special elections were themselves wrong, for example. It's all well and good to say the early election results don't look like the November polls, because we expect a different electorate in November, but they didn't look like the early election polls either! Were they polling those races based on the expected November electorate? That makes no sense whatsoever. If they were then that seems like a pretty glaring polling flaw.
The bottom line is the pollsters got those election results wrong. They can pontificate all they want about why November will be different, but no amount of theorising will erase the wrongness of their early election polls. It's quite simple, if they were wrong with these elections, they could easily be wrong about November's election too. That's not saying they will be, but this particular theory is not a hugely compelling argument that they won't.
How about the House and Senate?
So far we have mainly been focusing on the Presidential election, but we will of course also have the legislative elections in 2024. Once again, every seat in the House is up for grabs, along with several Senate seats. What do we expect here?
So far we have mainly been focusing on the Presidential election, but we will of course also have the legislative elections in 2024. Once again, every seat in the House is up for grabs, along with several Senate seats. What do we expect here?
In the House, Democrats appear to have the edge. For the last several years, Republicans held an absurd systemic advantage in the House due to gerrymandering, the process by which politicians choose their voters rather than the other way around. From 2010 to 2020, this was so bad that Democrats needed to win nationally by about 5% in order to take a majority.
However, following the recent redistricting process, and the broader implementation of anti-gerrymandering laws, this advantage seems to have all but vanished. In fact, in 2022 Republicans won the national popular vote, yet only managed a slight edge in the House, by 6 seats. 2024 is expected to be demographically more favourable for Democrats, on top of which Democrats currently lead in the national popular vote polls. Anything can happen, but on the surface it is difficult to imagine Democrats not taking the majority when they came so close in 2022 and seem almost certain to improve on those numbers in 2024.
In the Senate, the situation is very different. 2024 is an absolutely brutal map for Democrats. They currently hold a 1 seat majority, and need to defend seats in reliably red states including Ohio, West Virginia and Montana, as well as key swing states like Arizona, Pennsylvania, Michigan and Wisconsin. By contrast, there are no clear pick up opportunities for Democrats. The closest possibilities seem to be Florida and Texas, two states which are reliably Republican at the moment.
In order to hold their Senate majority, Democrats will need there to be some ticket-splitting. That is to say, they will need states that Trump is almost certain to win, like Ohio, Montana, Florida, Texas, etc, to vote for a Democratic Senator. Ticket-splitting is increasingly rare, especially in Presidential election years.
Democrats will be buoyed by the positive early polling in those states showing the Democratic candidate running ahead of Biden, which indicates that they could possibly hold those Senate seats even if Biden loses the state. The trouble is there is zero margin for error. Democrats will almost certainly lose the West Virginia seat now than Manchin has retired, which means they need to run the board on basically everything else. With margins this close, you would expect them to lose at least one, even if they are polling ahead. It's possible, but highly unlikely that they hold this majority.
Summary
Ok, so that is all for now. If you take one thing away from this piece, it is that we are still very early in the election cycle. We don't have a huge amount of concrete data, and there is good reason to believe that what little data we do have is prone to change.
It is far too early to draw any kind of conclusions about how these elections will go, but the baseline fundamentals favour Biden, along with much of the early data. There has been a lot of noise about the polls potentially leaning the other direction towards Trump, but under greater scrutiny that narrative appears not to be hugely compelling. It is very early and we will get much better data in the coming months. As always, this blog will publish a final forecast shortly before the election. Until then, remember not to get too caught up in the media's hysteria. Data will fluctuate up and down over the campaign, it always does. Not every single little data point is significant, even if it makes for a punchy headline.
