There are 20 days to go until November 3rd. Over 10 million people have already voted.
Early voting has begun in Texas. And just like Georgia, people waited in line for hours on the Lone Star State’s first day of voting. As you can see from these pictures, long lines were the standard yesterday.
In case you (or someone you are arguing with) thinks the pictures are selective and appear longer due to social distancing, we have data for how many people voted yesterday.
That’s right. One million people voted early yesterday in Texas. One million! In Houston, about 170,000 people voted yesterday. That’s 40,000 more than four years ago.
The current state of election forecasts
As we get past the three week out mark, it’s time to start looking at the forecasts. Forecasts and polls (including polling averages) are not the same thing. Polls are a snapshot in time of public opinion. Polling averages are a way to understand what various polls are saying about the same election over time, but they too are snapshots. Forecasts, on the other hand, attempt to project the result of the election.
Forecasts use polls, but often also include other information like historical trends and assumptions about how the economy (for instance) correlates with electoral outcomes. The three most important forecasters are FiveThirtyEight, The Upshot (New York Times), and The Economist. Other folks like Cook Political Report do a kind of forecast based on how they view the likelihood of candidates winning states or districts. CPR uses a lot of on-the-ground information to inform their state and district ratings (using the Solid, Likely, Lean, and Toss-Up categories). Others like Inside Elections and Sabato’s Crystal Ball do similar analyses, all of which are used to create the Consensus Electoral Map. The Map itself is a kind of forecast.
The big three forecasters are explicitly doing just that: forecasting the election results. The others are doing something similar, but remain focused on simply whether it is likely or not that a candidate wins a state. So, with the others we see simple electoral vote scenarios that are, in those outfit’s opinions, the likeliest. FiveThirtyEight, The Upshot, and The Economist attempt to predict the odds that candidates will win the popular vote and the electoral vote. It may seem like splitting hairs to distinguish the two types of forecasting, but the ones that most people have in mind when they think of election forecasting are the ones done by the big three.*
Many people seem to remember “the forecast” in 2016 was that Clinton was guaranteed victory. It’s the main reason why people still to this day think the polls were wrong and they cannot be trusted. However, what people “remember” is that The Upshot gave Clinton an 85% chance of winning. That means Trump had a 15% chance of winning, not a very good chance but not impossible either. Since FiveThirtyEight is the only forecasting outfit that regular folks think of, many people “remember” FiveThirtyEight forecasting a definite Clinton victory. It did not.
The final forecast for FiveThirtyEight in 2016 gave Trump a 30% chance of victory. That it is big number in probability terms. And at the time, Nate Silver said clearly that Trump was just a simple polling error away from winning. There is always polling error, so that was no criticism of the polling. In fact, the polling error in 2016 was pretty close to being standard. We remember the error because Trump won. In 2012, the polling error was bigger than 2016. We don’t remember that because the error was in Romney’s favor and he lost.
Periodically over the next three weeks we’ll check in with the forecasts to see how they are tracking. After the election is over we’ll revisit the forecasts and see how well they did. We’ll do this with the individual polling as well. On Friday in our usual weekly polling round-up we’ll take a look at the current state of things by quality, average, and mediocre pollsters. But for now, let’s check in with the big three forecasters.**
FiveThirtyEight’s current forecast is below. As of today, the chances of a Biden victory are significantly greater than they were for Clinton during the weekend before the 2016 election. However, Clinton’s chances were bigger earlier in the campaign so the forecast can change if public opinion changes. However, this year’s race has been so stable - and any change seems to go in Biden’s favor - and Biden’s favorability is significantly better than Clinton’s was that it would be an impressive change in public opinion if the race got close in the next 20 days. To be clear: it can happen.
The Economist’s current forecast by state is below. This type of forecasting is something that FiveThirtyEight and The Upshot also do, but they focus more on electoral votes than on state probabilities. In this instance, The Economist is employing a forecast more common with CPR, Inside Elections, Sabato, and others.
This forecast projects Biden winning between 229 and 421 electoral votes and Trump winning between 117 and 309 electoral votes. The probability is not uniform across all plausible votes. In fact, The Economist projects that Biden has a 92% probability of wining the electoral vote. You can’t tell that from the chart above, but The Economist lays it out to us clearly in the chart below.
The Economist is newer to the forecasting game than the others, but the work they are doing is very sophisticated. That does not mean they are better or more correct. In fact, both FiveThirtyEight and The Upshot (not the mention the other forecasters) are more cautious about Biden’s chances of winning both the popular vote and the electoral vote.*** Here’s how the Economist is forecasting the national popular vote:
What’s not completely apparent in this chart is that The Economist’s current forecast is for Biden to win about 54% of the vote and Trump to win about 46% of the vote. That forecast is very similar to the current polling averages (see chart below), but since the race has been so stable over the past six months it makes sense with less than three weeks to go. According to The Economist, “[t]he model first averages the polls, weighting them by their sample sizes and correcting them for tendencies to overestimate support for one party. It then combines this average with our forecast based on non-polling data, pulling vote shares on each day slightly towards the final election-day projection.”
Let’s check back with FiveThirtyEight to see what they are projecting for the popular vote.
As you can see, this forecast is almost identical to The Economist’s. Both currently project Biden winning between 53 and 54 percent of the vote and Trump getting between 45 and 46 percent of the vote. These projections will likely be accurate but not correct because they do not take into account minor party candidates, which are currently polling collectively at around three percent nationally.
The Upshot so far is choosing instead to stick with how the election would turn out today based on current polling averages. I am not sure if The Upshot will stick with that through November 3rd or go back to the projecting the relative chances of victory for Biden and Trump as it did in 2016. It appears the folks at The Upshot are a little embarrassed about last time’s forecast. One interesting thing that The Upshot is doing this year is demonstrating the difference in outcomes based on polling error rates from 2016 and 2012. It’s not in the chart below, but based on the 2012 error rate The Upshot projects Biden winning 375 electoral votes - the same as is projected without the error rate.
That’s the current state of the big three’s forecasting. Remember that forecasts are dynamic and will change with new information. There is no perfect poll or forecast. It’s best to think of them as targets that can err in either direction. In polls, consider the margin of error. In forecasts, consider the range of probabilities. In my opinion the best way to look at both is to consider who is winning and the magnitude of the margins. For instance, if the polling shows leads changing regularly the race is likely a true toss-up. If the polling shows one candidate consistently leading, he or she is probably winning. If the polling shows that candidate consistently winning by a large margin, you can feel more secure in projecting that candidate as the likely winner.
* I am using the term “big three” merely for simplification. It’s no reflection on the good work done by CPR, Inside Politics, Sabato, or anyone else.
** We’ll check in with the other forecasters later, and perhaps by then I’ll have better names for categorizing the two different groups of forecasters. No promises, however.
*** FiveThirtyEight gives Biden an 87% chance of winning the election and The Upshot has not even offered probabilities yet.