2020 Election Night Watch Guide

From the primaries to the pandemic, this has been a wild election year. November 3rd will (hopefully) bring an end to the chaos of the campaign trail. So what is likely to happen? I wrote this guide to help people better make sense of what is going to happen on election night. First, I start by briefly walking through the evidence that the polls are valid and generally should be believed. Then, I’ll talk about what the state of the race looks like before the night even begins. The rest of the article is then designed to take viewers through a chronological list of the states they should be paying attention to as election night winds on, and a look at the House and Senate.

On polling, 2016 and 2020

Should you believe the polls? Yes, you should. To briefly summarize a mountain of evidence:

1. Individual polls call elections correctly about 75-80% of the time. This is true of both presidential elections, and elections generally.
2. Aggregate models do much better than any given individual poll on average. FiveThirtyEight’s model, for instance, predicted all 50 states correctly and came within a tenth of a percent of the popular vote in 2012, and predicted 97% of races correctly in 2018.
3. These models were not far off in 2016, either. Comey’s decision to reopen the investigation into Clinton’s emails weakened her lead in the last several days of the race by several points, and turned a race heavily favoring Clinton into a toss-up that only narrowly favored her. The aggregate model missed Clinton’s eventual 2.1% popular vote victory by 1.5%, a small and typical polling error. The real problem came from polling in swing states like Wisconsin, Michigan, and Pennsylvania, and even in those cases, the polling errors tended to be within the margin of error.

Source: Washington Post

4. 2020 is not the same as 2016: the race has been unusually stable compared to 2016, with polls swinging just 3 points over five months of campaigning in 2020, compared to 12 points of swing in 2016 (see above); Biden’s lead is more than double that of Clinton’s at this point last year (also see above); only about 4–6% of voters appear to still be undecided in most recent polls, compared to 10–15% in 2016 at this point; and Biden has a significant favorability advantage over Trump, compared to Clinton, who was tied for the most unlikeable candidate of all-time with Trump by the end of the race.

5. Pollsters, who don’t want to be criticized again, have updated their model to account for 2016’s surprise turnout demographics. Four years ago, polls systematically underestimated the likelihood of white voters with no college degree turning out. Now, in 2020, they are re-weighting their respondents so that these voters are accounted for. In fact, to the extent that there is a polling error this year, I would expect it would be somewhat more likely to advantage Trump than Biden, and so we might instead be underestimating Biden’s performance this year — pollsters don’t like to be wrong, much less famously wrong, much less famously wrong in the same way two elections in a row. My suspicion is that polling agencies, probably have built models that would be somewhat slower to predict a Trump loss.

In short, when FiveThirtyEight gives Trump a 10% chance of winning as of today, and The Economist’s model gives him just a 4% chance, you should generally believe them. However, 90–96% is not 100%. Unlikely things happen all the time. All we get to say is that Trump’s victory would be very surprising.

These models estimate Biden’s popular vote to be about 8.5%, and that tells us something about his chance of winning the electoral college, as there is a strong correlation between electoral college margins and popular vote margins. A 2% margin of victory or less creates conditions for the electoral college playing a role. Clinton had a 2.1% margin of victory in the popular vote, and lost the electoral college by about 70,000 votes across three states. Once the popular vote margin gets above 3%, it becomes much more difficult to envision a scenario in which the electoral college would be relevant. To get to a place where things would be dangerous for Biden, the models would have to be off by about 6%. That would be an error of historical proportion. Such an error is possible, but very unlikely.

Calling Races

The high volume of mail ballots, combined with slowdowns in the postal service, means that there will likely be a large number of votes left to be counted after midnight in many or most states. In most states, this won’t matter, because the gap between the candidates will be large enough that networks will be confident in calling the election for one candidate or the other after just a small percentage of returns have come in. In the swing states, however, a close race could mean that virtually 100% of votes would need to be counted in order to declare a winner, and this could take days.

Another issue with mail ballots is they are likely to be skewed in terms of candidate preference. Democrats have indicated repeatedly in surveys that they are about twice as likely as Republicans to vote by mail, so mail ballots will probably tilt heavily towards Biden. This means that the vote totals on election night will lean quite a bit more towards Trump than they will after remaining ballots have been counted over the next couple days.

As such, news organizations will be more reluctant to call states for Trump than they will for Biden, as there’s a fair chance that an apparent Trump lead with 10–20% of votes left to count will be overturned eventually, while there is a very low chance of the reverse happening. If Biden wins in the landslide the polls predict, the election will probably be called for him at some point on the evening of November 3. If either candidate is leading by a narrow margin, especially Trump, it’s going to be a long, chaotic, and extraordinarily stressful week, weeks, or month of counting the remaining ballots and filing lawsuits.

Setting the Table

Election night will be easiest to follow if we filter out the things that don’t matter. The vast majority of states are not remotely competitive, so it’s not worth cheering your side’s every victory, and panicking over your every loss. We should focus only on the states that election models suggest are truly competitive. I start by awarding safe states’ electoral votes to their victors, according to the estimates of aggregate polling models. I use a threshold of a 5% polling margin between the candidates to determine these safe states, given how unlikely a greater than five point error would be.

Using this method, Biden starts out with an impressive 258 electoral votes. He only needs 270 votes to win the presidency. Biden’s 258 votes come from all the states that Clinton won in 2016, plus Wisconsin and Michigan. In Wisconsin, Biden is favored by about 8.5%, and in Michigan, he is favored by about 9%. New Hampshire, Nevada, Colorado, Minnesota and Virginia usually bear some monitoring as swing states, but polls indicate this year they are all highly likely to vote for Biden.

What if the polls are wrong, and these safe states aren’t so safe? Well, then you can go right ahead and scrap most of the rest of this article. This article is written for the 95%+ of possible universes in which none of those states are in play.

A brief note on Texas

This state has received a great deal of attention given that it is not usually competitive for Democrats, and now polling indicates it is. Still, you should mostly ignore Texas on election night. Outside of being a nightmare for Republicans, it won’t matter much for the results. Texas has almost no chance of being decisive; if Biden wins in Texas, that suggests he already has a landslide victory from winning many other states. It’s hard to imagine how Biden could narrowly win Texas while also losing in, say, Pennsylvania. Also, while Biden could win this state, Trump is ultimately still favored (+1.0). Finally, it’s a geographically large and populous state in the middle of the country, and given a close race, it’s very doubtful Biden would have the state called for him on election night. Let’s focus on more important states.

A State-by-State Watch Guide

The following states are arranged to aid viewing. You can start by monitoring the first states listed below early in the evening, and as the night goes on, move down the list as necessary.

New Hampshire: As a small state on the east coast, New Hampshire should count quickly and early, and therefore provide us with our first good hint about how accurate the polls have been. Polls predict Biden should clearly win, by about 10 points. If it looks like Biden is winning, but in a much closer race (say by 3–5 points), this could be our first indication that it’ll be a tight night for Democrats. If Biden appears to retain about a double digit lead, and New Hampshire is called very early, this indicates things will likely go poorly for Republicans.

The Key Six: Trump has to win six states — Pennsylvania, Arizona, Florida, North Carolina, Georgia, and Ohio — if he has any shot of beating Biden. If he loses even a single one of these states, he’ll be done for, barring completely absurd circumstances.

Pennsylvania: Of the six states Trump has to win, this is the least likely. Biden is up by an average of about 5 points in the state right now. A polling error of great magnitude could swing the state to Trump, but it’s fairly unlikely. Of the six states, this is the state that could get called in Biden’s favor first. If Biden wins Pennsylvania, the election is effectively over, as Trump no longer has a path to 270 votes.

North Carolina: If Pennsylvania looks close, and the news orgs aren’t calling it yet, this is now the next most important state to watch. Biden currently leads by about 2 points. If it’s too close to call in PA, it’ll probably be too close here too, but weirdness in sampling might give Biden more of an edge in NC relative to PA than expected. Like Pennsylvania, if Trump loses here, the race more or less is over.

Florida: Like North Carolina, Florida shows a Biden lead of about 2 points, but you’re more likely to see NC called before FL. Florida is notorious for administrative difficulties in polling, and also given its larger population, Florida may take longer to count than North Carolina. But like the other two, a Florida loss ends Trump’s presidency.

Georgia, Ohio: Hopefully, we won’t live in a universe in which these states come into play. If Georgia or Ohio is needed for a Biden victory, it’s going to be a very long and harrowing night, and probably week…or more. Ohio is more or less tied, and Georgia is narrowly leaning towards Biden by 1 point. Trump losses in either would end the election, but if we’re at this point, don’t look to these two states for an answer. Instead, you should look to…

Arizona: With a +3 Biden average, this state is actually more likely than any key state other than Pennsylvania to decide the election for Biden. However, because it’s a west coast state, we probably won’t see it called for one candidate or the other until much later into the evening. By this point, if the Biden camp isn’t already celebrating their victory, their best hopes will hang on what happens in Arizona. If Trump loses Arizona, then his victory is almost assured, but one more thing would have to happen if he wins the other five key states but loses Arizona…

Nebraska’s 2nd, Iowa: Biden would need to combine an Arizona victory with one of these two to secure his win. The more likely of the two is NE-2, which Biden currently leads by 5 points, and which would give him an extremely narrow 270–269 electoral college victory. Iowa would be much harder, as Trump currently leads in polling by about a point, but it would also provide a path to victory through Arizona. There’s an extremely small chance that Texas could give Biden the victory if Iowa and NE-2 did not, but this would be a genuine shock.

The House

Democrats are virtually guaranteed to win the House, quite a bit more so than the likelihood Biden becomes president. The vast majority of seats in the House are likely to be safe seats, so turnover is confined to a handful of small races. In all likelihood, Democrats expand their House majority from 233 to something in the 236–241 range. Nevertheless, there are still a few interesting House races to watch.

NJ-02, Van Drew (R) vs. Kennedy (D): Van Drew, the incumbent, was a Democrat before switching his affiliation to Republican after Trump’s impeachment. Kennedy is yet another member of the famous family. The race is a dead heat.

CA-48, Rouda (D) vs. Steel (R): Harley Rouda won back this Orange County district from long-time incumbent Dana Rohrabacher in 2018. Steel is a strong challenger, and keeping this district in Democratic control would demonstrate that the flip in 2018 wasn’t just a fluke.

AZ-06, Schweikert (R) vs. Tipirneni (D): Schweikert is a multi-term representative of suburban Phoenix. His loss would be emblematic of the problems that Republicans appear to be having with suburban voters this year.

The Senate

Of the White House, House, and Senate, the Senate is the chamber where Democratic victory is least likely guaranteed, but Democrats are still favored to win narrow control according to polls.

Likely Democratic Losses (1): Democrats are expected to keep almost all of their seats. Doug Jones (AL) is the only Democratic incumbent who is expected to lose on election night, given that he won largely because of the unpopularity of Roy Moore as a candidate, and Tuberville is a high quality challenger. The only other Democratic incumbent race worth paying attention to is in Michigan, where Peters currently leads by about 7 points. An upset here would be very unlikely, but possible.

Likely Democratic Gains (4): Two Democratic challengers are likely to win their races, Hickenlooper in Colorado, and Kelly in Arizona. Colorado is a purple-turned-blue state, and Arizona a red-turning-rapidly-purple state. Both Kelly and Hickenlooper are high quality challengers, and currently favored by 6–7 points in their races. Assuming the night appears to be going as well as expected for Democrats, two more are Democratic pick-ups are lean likely: Cunningham in North Carolina, and Gideon in Maine. These two challengers are expected to beat Tillis and Collins respectively, but they are favored only in the 2–3 point range, so a moderate polling error in Republicans’ favor could put them into toss-up status.

Toss-Ups (2): There are two races which appear to be dead heats — the Greenfield/Ernst race in Iowa, and the Ossoff/Purdue race in Georgia. Both Democratic candidates have made late gains in the polls, and now have roughly even chances of winning. These races probably won’t get called until after election day, given how close they are likely to be. The Georgia race, given Georgia election rules, could even go to a January 5 runoff if the Libertarian candidate prevents both candidates from getting above 50%.

Control of the Senate: At this point, if the likely seats go as expected, and we assume Democrats win one of the two toss-ups, Democrats will have 51 seats and majority control of the Senate. If a polling error favors Biden, additional seats will be in play.

Not-so-Safe Republicans (3): Three Republicans are currently expected to win their races, but have a narrow lead that is vulnerable to a blue wave — Graham (SC), Daines (MT), and Loeffler (GA). The first two are potential grabs in the case of a clear Biden victory on election night. In the latter race, Democrat Warnock will likely receive the second-most votes out of multiple candidates running, which will put him into a runoff against one of two Republicans on January 5. Special elections favor Republicans, given that Democratic turnout is typically lower in these races, so a Warnock victory would be fairly difficult. But if all three of these were to tip blue, the toss-ups would also probably both go blue, and Democrats would have 55 seats in the Senate. This would be around the same odds that the polling error is so high in Trump’s favor that he wins the election.

Blue Tsunami Republicans (3): Although it is very unlikely, if there is a truly sizable poll error in Democrats’ favor, Cornyn (TX), Sullivan (AK), and Marshall (KS) would all be in danger of losing races in which they are heavily favored. In Republicans’ worst-case scenario, Democrats would end up controlling perhaps 58 Senate seats. Again, this is extraordinarily unlikely, but it’s getting 57 seats, for instance, would be about as likely as Trump winning the popular vote. We should consider extreme possibilities on both sides of the distribution.

Conclusion

Though Trump’s victory is highly unlikely, election night might feel like it’s going well for Trump, given the partisan disparity in mail-in voting. I close with four general scenarios for the night and beyond based on FiveThirtyEight’s model. Each scenario’s title refers to the eventual outcome after all ballots are counted, and not what will happen on election night:

Biden Landslide (1 in 4 odds): Biden will probably appear to lead modestly throughout the night, and the election will be called for him that night or in the early morning.

Biden Wins by Predicted Polling Margins (1 in 3 odds): It might look close throughout election night, but Biden would have the election called for him in a key state like Pennsylvania, Arizona, or North Carolina early in the morning or next day.

Biden Wins in Close Race (1 in 3 odds): Trump will run close or lead in many of the key swing states throughout the night, and panic will set in amongst Democrats. Trump may try to declare victory at some point before networks do. Counting will likely go on into the next days or even weeks. Lawsuits will be filed, and distrust will be sown. This scenario would by far be the most dangerous to democratic norms and leaves room for some very awful possibilities.

Trump Wins (1 in 10 odds): Trump will appear to have a commanding lead throughout the night. The races will tighten, but networks will start considering whether to call the race for Trump or not. Perhaps some will. Mail ballots will be counted over the next several days with a growing sense of dread for Republicans that it won’t be enough to make up the difference. There’d be a few recounts, but Democrats would probably accept their fate by mid-November, barring clear evidence of administrative error or corruption.

See you all on the other side!