Tom,
I had the thought of writing to you about how I engage with election forecasting based on polling data. I thought (presumptuously) you might be interested in a dive into the way I utilized the fivethirtyeight.com election forecast this year. I trust Nate Silver/538 forecasts more than other options because they have an excellent track record and I like what I know about their methodology (not that I understand the math). I followed their blog on Election Night 2016 and knew several hours before the networks started calling it that Trump won.
How I work with the forecast is not all that complicated and I tried to lay it out clearly. Of course, you will already know a lot of what I’m discussing and I wouldn't blame you for not wanting to wade through the discussion, but in any case, I found it helpful to explain what I do to myself. We often have complex habits that develop organically without a written plan.
2020.
Here’s the link to their final forecast. It was posted on Monday, Nov 2 and is their last word, written in stone (i.e., it won't change), allowing anyone to critique their work after the votes come in.
538 uses all reputable polling data available from many different outfits. With those numbers, they create a mathematical model factoring in myriad demographic variables, historic partisan trends, the state of the economy, and various other factors to calculate odds on every state going red or blue (see forecast). These folks are serious data nerds! In their big mathematical model, they average these polls but weigh each one differently based on their assessment of the quality of the methods the various pollsters use.
Quantifying and crunching all those complex variables by running 40,000 simulated elections where numbers vary under different scenarios such as high turnout, low turnout etc, 538 assigned Biden an 89% chance of winning the election. That is, in 89% of the simulations they ran, Biden won.Their results from the same simulations gave a Biden landslide a 30% chance, a Trump win 10% and an Electoral College 269-269 tie 1%. (A few days ago, a tie would have resulted had Georgia gone to Biden and all the other uncalled states, AZ, PA, NV, NC and Alaska, to Trump!) The remaining 59% covered outcomes ranging from a squeaker for Biden to a solid Biden win short of a blowout. None of these predictions was a certainty because there’s no such thing as certainty in the forecasting game, whether it’s elections, sports or the weather. The odds just give a sophisticated guess of what is likely to happen based on mathematical data. However, it's worth noting the actual result (from 295 to 306 EV, depending on AZ) will fall within that 59% range, that is, the range Silver predicted as most likely.
As you know, political pros use polls in Presidential campaigns to decide where to focus effort and allocate resources based on a strategy to get to 270 EV. We saw how the small set of states that weren’t called within 24 hours (except maybe Alaska) were all rated as close calls. Based on the polling each campaign did, those states were all targets of intensive efforts in the final days and weeks up to Election Day. The point is, the pros believe in their polling data and put their money on the line accordingly. Turnout for both parties was massive leading to long days of checking and counting the ballots; and, all of them were, in fact, close races with winning and losing hanging in the balance.
How did my go-to operation (538) do on the Presidential election? They were correct on Biden because he won (as has become clear now). They correctly called every state that Trump held from 2016. They correctly called every state that stayed blue from 2016. They correctly called Pennsylvania, Wisconsin, and Michigan to flip from red to blue. They correctly called the districts in Maine and Nebraska that can go either way. They called Arizona and Georgia to flip red to blue- the outcome of those two is somewhat in doubt so we’ll see. They missed on two big ones- North Carolina and Florida but in both cases they had rated Biden only a “slight favorite.”
In their final national forecast, which you can browse at the above address, you will see a graphic called the “snake chart” that shows all the states lined up in order of the 538 calculation of the probability of each one going red or blue. To fit it into a frame, the line of states twists around (sort of like a snake). At the extreme red end is Nebraska’s 3rd district followed by West Virginia and Wyoming. At the extreme blue end are Washington DC, Vermont, Hawaii and Massachusetts. These are easy calls that no respectable forecast would miss. You and I could make them every time if we knew how they've voted in the last 10 or so elections.
In the middle of the “snake” are the states and districts that are toss-ups. They are rated blue or red but they are the most uncertain according to the forecast. In this range, the last 5 in blue are Arizona, Florida, NC, Maine District 2 and Georgia. The first 5 red states in this range are Ohio, Iowa, Texas, Montana and South Carolina. If you start at the blue end and add the Electoral Votes (EV) for each state in order of the 538 probability of a Biden win (highest to lowest), the snake chart shows Pennsylvania as the “tipping point” state, the one that puts Biden over the top, because adding it’s EV to the ones that came before gives Biden 270 or more, enough to win. 538 assigned an 85% probability of Biden winning PA. The states that come afterwards would give Biden extra EV beyond 270 if everything came out according to the forecast. (A final snake chart after all results are in ordering the states from largest margin of victory to smallest is used to define the "true" tipping point state.)
The reason Biden got such a high probability of winning (89%) is because he has a lot of states past the tipping point, meaning there were more different ways he could win in the simulations than Trump had (based on their estimations, of course). Interestingly, the snake chart shows exactly the way it has turned out- all the states on the chart from DC through PA voted blue and Biden won. So as far as the Presidential race went, that’s pretty impressive to me. As a fairly sophisticated student of the electoral map, if forced to guess I would have said Arizona blue (right), Florida blue (wrong), no clue on Maine 2 (wrong) and everything else red (i.e., wrong on GA). Iowa struck me as a possible upset red to blue (wrong). So, I was wrong on FL, NC, GA and IA. Along with Maine 2 left blank, I got 50% right, as compared to Silver getting 80% of the close calls right.
But why did I care about the predictions? I knew they weren’t certainties and I’m not running a campaign. I just wanted Biden to win because I think Trump is causing great harm to the country and the entire world. You read my “Democratic platitudes” (and I realize some people don’t see it that way.) Why not ignore all that and just sit back and wait for the election to be called?
The answer: I used them to prepare myself for what was likely to happen, to make sense of the vote as it was reported, and to discuss what’s going on in real time with my friends in the liberal support community called Owen’s Geo-Political Analysis Group and elsewhere.
Since any number of scenarios are possible, I used an interactive map like
this one (OK, I used this exact one but there are lots of them out there) that allows you to create different outcome scenarios to look at different ways Trump or Biden could win. Hours and hours of fun!
What did I conclude? First, that if Biden got either Florida or Pennsylvania, he would almost certainly win as long as he also flipped Michigan and Wisconsin and held Minnesota. This looked promising. Second, I saw that even if Biden lost both PA and FL, he could still win by flipping Arizona, Michigan, and Wisconsin and holding the 2016 blue states (which include Nevada). Maybe Biden could pull an upset or two, as well (didn't happen). I wasn’t counting on Biden winning Georgia but I thought North Carolina and Arizona were likely. Getting one or more of those three would boost Biden’s different ways to win. That’s basically what I saw. I was pretty confident going into Tuesday evening.
As the results came in on Tuesday, it was quickly clear that Trump won Florida. Trump also went well ahead in Pennsylvania and other critical states on the strength of same day, in person voters, not a surprise since he urged his base to vote this way. He also made the false claim that voting by mail would result in massive fraud. Why did he do that? Because his polling pros told him this would make him appear to be winning the election. This would allow him to falsely claim victory so that if the well known "blue shift" caused by counting mail-in ballots occurred, he could say, "See, I'm being robbed!" which he did and continues to do as of this writing.
So by Tuesday evening it was clear the counting in PA would take a long time and Biden had a good shot of catching up and, hence, winning: Millions of same-day PA votes were still not reported from big urban districts and millions of mail-in ballots were going to be counted last. I went to bed at 1 am because the outcome was obviously going to take at least several days to determine. Before I did, I noted that an intelligent MAGA friend of mine had posted that it was incredible how Trump was going to win the election by huge margins, proving the dumbass prognosticators were totally wrong! Another knowledgeable friend and I responded to point out very politely that, because of the known character of the outstanding votes, it was quite possible, if not likely, the situation would get close and even reverse overnight The next morning, she and I had a private message from him saying we were right and he had deleted his post. We’re all still friends.
Still on Tuesday, I looked at other states to consider Biden’s chances assuming PA didn’t flip to blue. As you know, over the course of the Tuesday and Wednesday, Michigan and Wisconsin went to Biden (as 538 predicted) and Minnesota held blue (as predicted). This reassured me that the 538 model was doing OK. Biden also built comfortable looking leads in AZ and Nevada, so his chances still looked good (recalling that Biden could make 270 with AZ, NV, MI, MN and WI without PA). I was a bit nervous but not discouraged by the overall picture, knowing further vote counting could change things.
On Thursday, as PA started counting the mail-in ballots, Trump’s lead shrank every time a batch was reported. I tracked this on a handy chart in the online New York Times. Informed analysis of the mail-in ballots from multiple reputable news sources indicated they would favor Biden heavily and be sufficient to overcome Trump’s lead and build a solid lead for Biden. This kept on happening and at that point, I quit worrying completely because it was virtually certain Biden was going to win by carrying PA. (My close friends and wife were afraid to believe this when I told them.)
The interesting question then became how many other states would Biden win and, above all else, would he carry our beloved state of Georgia? Even though 538 rated Biden a very slight favorite in Georgia (58% win probability) before the vote tallies started coming in, it was very hard for me to believe a Democrat could carry my home state. However, when I examined a list of Georgia counties on the
New York Times webpage showing what percentage of votes were in and what % had gone to Biden or Trump for every GA county (a long list!), I could see that the outstanding votes were mostly from large, heavily pro-Biden counties such as Chatham and Clayton. To my surprise, it looked to me like Biden had a very good chance of catching Trump and taking the Peach State! This was confirmed by the analysts on 538 and other sources. I also noted that, since Biden was winning and Ossoff was losing, a lot of voters must have voted for both Perdue and either Biden or Jorgensen.
As of this moment, Biden is in the lead by a small margin of ~11,000 votes. That’s still too close for the news folks to call given the automatic recount that will be triggered. But Biden has it and I am impressed that 538 gave Biden a slight advantage that has panned out.
AZ is also looking good for Biden here on Monday evening.
So, you can see that for me, anticipating what’s going to happen is not simply a matter of seeing which candidate has the highest number for each state and waiting to see if that’s how it comes out. It’s a process of looking at multiple scenarios involving the states not all coming out like 538 predicted they would and following the unfolding vote count in light of substantive considerations such as which areas of a state have reported, how their votes are running, whether a state started counting early votes prior to Tuesday, what order a state is counting it’s different types of ballots and other factors. The 538 election live blog kept me infomed of developments on a timely basis. Because I had studied the possibilities closely, I could make sense of things as they happened and maintain a rational hope for a Biden win throughout. That was very helpful in keeping my anxiety level in check and providing realistic hope to others who dreaded four more years of Trump. It's also kind of fun.
Now, how well do I think 538 did overall?
-On the Senate, not so good. The forecast said a 75% chance of Dems taking the Senate at least 50-50, which is very unlikely now unless they win both Georgia run offs or all the outstanding ballots in Alaska are for the Independent candidate. They correctly picked red to blue flips in Colorado and Arizona, missed on North Carolina and Maine but did correctly predict Peters to hold his seat in Michigan (whew). They also predicted Graham to hold his seat and McConnell to crush his opponent, which he did, and Jones to lose to a former Auburn football coach. I know someone who is pleased with these results and I can certainly live with them, given Trump being fired as President. Of course, we still have the excitingly crazy Georgia Senate runoff to look forward to. Again, my emotions say the Dems will lose but we'll see.
-On the House, a below average forecasting outing (the House will still be blue but not as blue). House races are harder to predict accurately because they involve smaller areas with less quality polling data available, so it is what it is. I wasn't concerned much about the individual races and didn't track any of them.
-On the Presidential race? Good to very good. They correctly picked Biden to win and have only missed two calls so far. They probably hit on AZ and Georgia, too, which I would say is a very fine record. Not as good as when Silver picked the winner and every state correctly in 2008 and almost all in 2012. Much better than 2016 when he gave Hillary a 2/3 chance of winning and missed key states flipping to red. But the margins of Trump’s critical wins were very small and Silver warned his readers that a normal error that went Trump’s way could bring him the win. Because an election forecast is about probabilities, not certainties. People may say the 2020 Presidential forecast was less than stellar but it worked well for me.
Regards,
Owen