The US General Election 2020-2021: My Predictions Reviewed

The overview

The US election is now done and we can confirm one very important point. That the vote counting in the USA is properly screwed up - please see this Tweet reporting ongoing counting in a New York House district on the 23rd of December. As a reminder the vote was on the 3rd of November. Sunderland it is not.

However we also learned that my predictions weren’t awful but were also not very good - in 2012 I won £120 from my bets and predicted every state without too much trouble, my 2020 record is less impeccable.

EDIT - 8th January: Please note that all this was written before Trump’s attempted (and deeply amateur) coup d’etat and whatever the fallout from that will be, this was a shocking event and it was omitted from this post only because it hadn’t occurred when I wrote it. With regards to timings Decision Desk had called the Georgia race several hours before the major networks, who called it in the middle of the disruption.

The presidency

Here are two predictions I made - the following is “as expected”, an expected typical result.

I also made a “bad day” map:

And the real results ended up somewhere in between:

Overall from my “as expected map” Biden underperformed, losing three states (Iowa, North Carolina and Florida - the first of which wasn’t even close) and one congressional district (Maine’s 2nd). I didn’t predict margins but had I done I would have been overly optimistic on the extent of his success in Michigan and Wisconsin, both of which were narrow wins, though Pennsylvania and and Minnesota were more in line with my expectations.

However Biden didn’t quite sink to the level of the “bad day” map - he retained Nevada and made his expected gains in Nebraska’s 2nd district and Arizona, as well as gaining Georgia - expected to be more of a stretch than it actually was. Ironically his electoral college total exactly matches that of Trump in 2016.

My prediction was fairly conventional one based off polling and I expect few people expected this map, though it was certainly a possibility ahead of time. However I doubt many people, asked to pin down a single map, would have correctly predicted that Georgia would flip to the Democrats but North Carolina would not.

Outside the headline results it’s worth mentioning that many trends evident in polling before the election were borne out in the actual results - cities and, notably, suburbs trended very heavily for Democrats leading to success. There were trends running against this, Biden suffered very poor results in the countryside for instance, and a collapse in the Hispanic Democratic vote also occurred.

Ultimately this is a reminder that these electoral college maps can hide significant details - Biden’s result doesn’t look that different from Obama’s 2012 (or even 2008) maps but the coalitions each party holds together have changed a lot within a single decade. In 2008 a blue North Dakota may have seemed more feasible than a blue Georgia.

The Senate

This is where my predictions fall apart somewhat. This was my worst-case scenario for the Democrats, based mostly off polling data.

Overall this map is mostly accurate, though with some narrower margins than I expected, with the exceptions of Maine and North Carolina where the expected Democrat gains failed to materialise and the two Georgia races which are still ongoing. And, as a reminder, 50-50 means the party with the presidency has control - with the vice-president being an extra, tie-breaking vote. Somewhat like how in English local councils the mayor can break ties.

The former of these is very surprising but altogether not the worst possible result - Maine’s Susan Collins built her reputation as a moderate and only faced a competitive race due to some votes that risked that reputation. It’s likely she’ll be happy to work cross-party to reburnish her credentials. So, as surprise losses go, this is probably the least worst one possible.

North Carolina was narrowing in polls but I do admit that I thought at both presidential and Senate level that a very narrow Democrat win was the worst-case scenario. In both cases I was wrong.

Georgia was more competitive for Democrats than would have been expected given results elsewhere - the conventional wisdom was very much that it was less likely than North Carolina. As such this worst-case map was drawn expecting Republican David Purdue to defeat Democrat Jon Ossoff outright in the first round of the election (ie 50%+ of the vote) and for Raphael Warnock, the Democrat in the special election, to ultimately lose his run-off.

This first worst-case scenario did not come to pass and at the time of the first draft of this post it’s currently voting day for two run-off elections with the Democrats holding an edge in early voting demographics and polling, though with the punditry slightly favouring a Republican win.

Of course at the time of publication and editing I already know the results of these but, purely to generate artificial suspense, they’re going to wait until the end of the article. With who controls the US Senate at stake. Will reality be worse than my “worst-case” prediction? Spend literally a few seconds scrolling to see.

The House

I had expected modest gains for the Democrats. This did not occur and they actually went backwards somewhat, though still retaining control.

How were my results so off?

To put a long story short the polls were off again. Despite popular perceptions this isn’t the norm and I have put my money where my mouth is on that one, and returned with a profit. So, as usual when this happens, we need to ask why.

Firstly we can rule out that they simply didn’t fix mistakes from 2016. In 2018 polling companies predicted the mid-terms far more accurately, even slightly overcompensating the extent of Democrat success in that election. Clearly they’d made adjustments that were proved to be effective, such as weighting by education.

However we can’t ignore that by 2020 the results were wrong again. Even the national polls were outside the margin of error (typically 3%), with Biden winning by 4.4% versus an 8.4% average lead in the polls - a result that, incidentally, shows how flawed the US system is as Trump’s 2016 popular vote loss led to the same electoral college outcome. District level polling, so useful in 2016, also failed to deliver.

State polling, by contrast, was highly variable. In Georgia Biden enjoyed a 1.2% lead in the polls, seen as unlikely by many, which translated into a 0.23% win - statistically that’s a bullseye. By contrast in Wisconsin Biden had an 8.4% lead in the polls which translated into a 0.63% win, which is an absolutely terrible result. These are the extremes but broadly polls were quite unreliable in the Rust Belt and patchy but with bright spots in the South. Occasionally an individual poll stood out among a poor average - in Iowa for instance the polling average was very far off but with the Iowa-focused Selzer providing individual polls that matched the end result closely.

Two notable states in the US South that also had inaccurate polls were Florida and Texas - in both cases the result for the Democrats was approximately 5-6% below what polling implied and tends to be attributed to their missing a large decline in the vote for Democrats among Latinos, who tend to be difficult to contact for polling. Results in Florida suggested initially that this may be focused on Cuban-Americans and Venezuelan-Americans, demographic groups that, understandably (and particularly in the first generation) may have very different politics to Mexican-Americans or any other “Latino” group. After all, Latino is not a sensible and coherent ethnic group - there’s a reason why the rest of the English-speaking world, or for that matter Latin America, doesn’t use the term as such.

Unfortunately for the rant I was going to write about how US use of the term “Latino” is socially regressive and is an extension into the modern day of the xenophobia encapsulated in historic American usage of terms like “white” as implicitly excluding groups they found undesirable at the time, such as Italian-Americans, post-election analysis showed this shift actually was found among Latinos in general. Oops. I guess I can make that rant and then make a begrudged counter-rant about how regardless of the logical validity of a term group identities are formed purely by people identifying with a term collectively no matter how illogical? Or maybe I can spare you either?

Elsewhere it has been observed that Trump drove up turnout in conservative traditionally low turnout groups, a similar effect to the Brexit vote closer to home, and that due to partisan distrust of the polling industry and high enthusiasm among Democrats (as well as an increased likelihood to be at home shielding from the coronavirus) the samples used for polling ended up far too Democrat-friendly.

This could be quite difficult to resolve or it could simply resolve itself with Trump no longer being on the ballot. Come 2024 we may know for sure.

And if you don’t like that explanation, I have others.

My senate races of interest

Hey, look, it’s this picture again.

In my initial post I’d mentioned several states of interest at the Senate level and thought it worth following up on them - these broadly fall into a few categories: expected results, unexpectedly uncompetitive results, subtly surprising votes and completely surprising results.

My expected results were Kentucky, Texas and Alabama - easy Republican wins, as well as Arizona - a solid Democrat win.

My unexpectedly uncompetitive races were Iowa, South Carolina, Kansas and Montana - these were all races I expected to be quite close but which the Republicans easily won. Many of these senators could have been real assets to the Democrats in the future and lost to fairly generic or lunatic-wing Republicans so these are disappointing results to me.

Alaska was arguably also in this category, an easy Republican win was certainly the default expectation, but the interesting scarce polling made it look potentially more competitive than it was.

My subtly surprising results were Michigan, Georgia and Mississippi. Michigan was far narrower than it was expected to be, as was the case in the presidential race but more pronounced, and Georgia was a marginally better result for Democrats than I think the data implied in the first round. Mississippi in terms of the actual result was pretty much as expected but this also hides the extent to which Mike Espy outperformed Joe Biden across the state - despite a decisive loss Espy was probably one of the best performers in the 2020 elections.

Finally my completely surprising results were North Carolina and Maine. In the former case the polls had narrowed but I’d expected a narrow Democrat win as the worst realistic result for them, rather than the loss they faced. With Maine I had thought Susan Collins a dead woman walking, politically, though in the end she pulled through (another big polling error!) in probably only her second competitive election in a long career. As I’d said before this is probably the Democratic loss I’m least sad about, though as a Briton who isn’t very right-wing indeed I’d have still preferred a Democrat flip in Maine.

Georgia run-offs

The end of the 2020 election actually occurred in 2021. These the runoffs decided control of the US Senate, and with the early vote favouring the Democrats going in were expected to be competitive in this traditionally Republican state.

The day before the election there was one last Trump scandal, with Trump insisting to the state’s Secretary of State that he had won and requesting that votes be “found” to show it. Given what Trump’s supporters have ignored in the past I don’t think this had any decisive effect, even in a close race, but it probably did draw press attention away from whatever the Republican candidates had to say - so chalk up another case of Trump hurting his own side for no clear benefit to himself.

Ultimately there was a long night of counting, with a strong initial Democrat lead whittling down to a narrow Republican lead before urban and absentee votes came in - would this be enough for a blue victory? I, good reader, valiantly slept through all of this. Anyway, the Democrats won narrowly, with Warnock’s victory being slightly but notably more solid than that of Ossoff.

And with that, the 2020 US election is finally over, with the Democrats narrowly controlling the presidency and both chambers of Congress. Congratulations to the winners of this tight race - Raphael Warnock and Jon Ossoff!

Now what?

For US politics this is probably it for a while, their politics tend to be quite dull outside of the set-piece elections to be honest so I’ll likely lose interest until 2022, and possibly even 2024.

However I’m intending to commit to at least one non-political project on this blog, I’ve done all the prep work for it so that’ll start whenever I fancy writing it up.

There’s also always elections going on so I may cover some things if I have time and inclination. There are many interesting European elections that I may make predictions and/or post-mortems for - for a start the Netherlands is usually good, Moldova’s often interesting and obviously Germany’s a huge deal. Beyond this there are almost ten other sets of elections across the continent that may be interesting enough to cover if they catch my eye at the right time. Finally there’s a mega-local election in the UK (including postponed 2020 elections) - I’m mildly curious about whether COVID-19 will push these into 2022. If not I may well cover them after the fact.

Outside of Europe the pickings are a little slim so far as interesting elections go - Hong Kong, if allowed to be free, may be interesting and Israel’s always a confusing, exciting mess when it comes to election time. Otherwise in Asia we only have the Tibetan government-in-exile elections (more interesting for their existence than anything else) and the Japanese general election. Japanese politics actually are interesting but, as most election nerds can tell you, their general elections are usually very dull affairs, typically voters being annoyed with the Liberal Democrats (no, not those ones) but voting for them anyway.

Elsewhere Africa and the Americas offer up even less - in Africa Somaliland (not Somalia) is holding its first election in over 15 years, again something interesting mainly for its existence in the first place. In the Americas we have the Chilean general election, which may be interesting, but otherwise the elections that most catch my eye are the Mexican and Argentinian legislative elections. As both have presidential systems these aren’t the “main event” though. If you like really confusing politics though Argentina’s always got your back.

So none, some or all of that may be coming down the road. But I’ll be switching off on the USA for a while.