Quick Note: At the time of writing, I was deciding on whether to do more predictions, and this article concludes on that question. However, I decided against doing more predictions in the 2022 midterm despite good data availability. I decided not to change the original post as an editorial choice to preserve the right tone.
Two years ago, I wrote Presidential and Georgia U.S. Senate runoff election predictions. It was for family and friends, and I didn’t initially plan to publish them. But when I needed more blog launch articles, I realized I could review these predictions for quick and good lessons.
Of course, learning from predictions can be complicated. It can be filled with hindsight bias, overconfidence, discounting the positive, oversimplification, self-serving bias, narrative fallacy, and who knows how many cognitive biases and distortions. But I’ll avoid as many potholes as possible because, honestly, I got the overall outcomes correct. 😊
What I Like and Didn’t Like About My 2020 Presidential Prediction
Let’s address the insurrectionist elephant in the room. I couldn’t have predicted January 6th. I sometimes wonder why I don’t see the depths Trump and his orbit will go to take power and rule as authoritarians. Then, I remember I wouldn’t naturally stoop that low which is why it’s hard to see those events coming. I’ll continue to sharpen my analytical skills regarding the Republican Party’s threat to democracy. But I rest well knowing that my personal default is democracy, equality, and freedom. Not kleptocracy and demagoguery.
Presidential Prediction Likes
Overall Outcome, Georgia, and Arizona: Well, I’m happy to get the overall result correct. President Biden is now president, and, while he has his work cut out for him, he did win. <insert sigh of relief>
I liked my non-strategic strategic assessment. I worked analytical muscles in new ways to assess suboptimal strategies by both candidates. A friend inspired me to do this when he pointed out the ridiculousness of using traditional political strategy to assess an election with Trump in it. He was right, and I was better off for listening. I also think it added to the strategic toolkit that I bring to work each day.
I also feel good about my Georgia and Arizona calls for Biden. While I didn’t write it for publishing, I researched whether I should even consider Georgia a battleground state. Once I believed that was the case, I had to get over the fact a Democrat didn’t win statewide in the prior cycle. That’s usually a good sign a state will flip. But Stacey Abrams lost under the haze of subpar strategy and alleged voter suppression. Arizona was easier to pick because Sen. Kyrsten Sinema won in 2018.
Quick side note: As a Biden voter, I felt equal parts good, and guilty, about picking Iowa and Ohio for Trump. But I followed the data, and that’s where it took me. Specifically, Ohio has voted for every presidential winner since 1960. This made it tough to give Ohio to Trump but give Biden the overall win. Again, I researched whether Ohio was even still a battleground state. Needless to say, my thoughts about Ohio and Georgia’s battleground state status flipped because following the data and my own election frameworks took me to interesting places in 2020.
Coronavirus as issue #1 and Coronavirus-based Models: Building a strategic framework focused on an election’s most important issue is a pretty standard prediction tool. Locating “Issue #1” in 2020 was both easy and difficult at the same time. Pre-pandemic, Gallup struggled to find a single issue #1.[1] After the pandemic started, Pew Research Center had similar challenges.[2] But the pandemic as the source of health, safety, and economic concerns had to be issue #1. I then used the importance-performance model, a business school framework, to assess how states might swing. If coronavirus was of high importance and Trump’s performance rating was low, then Biden would likely win the state. This method produced a very accurate map and revealed Biden’s overall strength. To get election predictions right, you must have a good sense of what narratives are moving the electorate even if you must look beyond the polls, media, and candidates to get there.
Using the Polls: 2016 made people skeptical about polling. Some people wrote polls off forever afterwards. I thought this was an extreme approach, and I decided to just get better at using them. FiveThirtyEight (“538”), the Cook Political Report, and similar sites give us incredible insight into pollster, methods, polling aggregates, and election dynamics. Through this, I got pretty well tuned in.
I also focused more on polling errors this time and used them to build multiple maps. I needed some intuition on how big and how likely an outcome-changing polling error would be. Using a high-level method from 538, I could only produce one winning Trump map and that helped calibrate my confidence in the polls. In the case of Texas . . . maybe too confident. There’s an art and science in working with polling and survey data, which requires applying additional analytical rigor to them.
Presidential Election Dislikes
Texas, Florida, and North Carolina: As hinted above, Texas really stings. My coronavirus-based framework had Texas red twice. In retrospect, I should have stuck with that. At the time, I felt I had picked Georgia for the same reasons that would apply to Texas, and it would only take a small polling error to make a Texas prediction correct. But with Georgia, I had more polls with Biden winning, high-level turnout math based on early vote data (thank you U.S. Election Project), anecdotes, and personal experience. I had none of this for Texas. I basically had an article about high early voter turnout in Texas.[3] I also turned my attention to Texas very late in the election and really should have used my lack of data as a reason to be conservative in my prediction. In writing this, I briefly looked at the reasons why Texas didn’t follow Georgia.[4] But I’m going to get to know Texas better at a later date.
To be frank, I don’t really think too much about missing Florida and North Carolina. No model for making predictions is going to be perfect, and 538’s model with more staff and data also missed those two. Overall, I was basically three states and some congressional districts away from a perfect map. Maybe with more time and data I would have gotten those too, but I’m never one to let perfect be the enemy of good.
History Makers Section: I called out Georgia and Texas as history makers, and, of course, Georgia went on to make history. But Arizona also made some presidential history. The state hadn’t gone Democrat since 1996. As mentioned above, Ohio also had a long streak of picking the winner broken in 2020 as well.
Trump Path to Victory Section: I did two things in this section that I didn’t like. First, I very casually threw out some democratically subversive tactics as a Trump victory path without covering how terrible for democracy that would be. As I said above, I had no clue January 6th was going to happen, and it is hard for me to imagine such depravity. But, in a post-January 6th world, I will no longer casually discuss subverting democracy. Second, and much less important, I didn’t write about the Republican data advantage and Trump’s appeal to low propensity voters. Both could contribute to a surprise victory, and I should have explored it more thoroughly. By the Georgia runoff, I corrected that error.
What I Like and Didn’t Like About My Georgia Runoff Prediction
For any other election year, this blog post would end here. However, with 2 seats needed to control the Senate after Theresa Greenfield (IA), Sara Gideon (ME), and Cal Cunningham (NC) all lost, the 2020 Georgia Runoff had high national interest. I, therefore, felt the need to write another friends and family prediction.
Like- Overall Outcome and Mathematics: Once again, I’m happy that I picked the winners. I did a considerable amount of math for this election, and I think that impacted my writing tone in that prediction. But the math did help me find the right result again.
I had quite the Excel workbook behind this prediction. Thanks again to the U.S. Elections Project and Dave Leip’s Atlas of U.S. Presidential Elections. To create multiple election scenarios, I had to use voter turnout as my model lever and set win numbers for the Democratic candidates. It was clear that turnout would drop from the general to the runoff and, with those win numbers, I could track the drop off statewide in the early vote. From there, I approximated what Republicans needed for Election Day and qualitatively assessed the likelihood of their party winning that turnout. This showed a narrowing Republican path to victory, and I became confident enough to pick now-Sens. Warnock and Ossoff to win.
One small nitpick, I don’t really remember what made me believe turnout would be on the lower end of my range. There were almost 4.5 million voters who turned out. I projected a range of 3.8M to 4.23M votes. I had models showing a high-end turnout of 4.7M, but I got conservative in my approach for some reason. Oh well, my final models had a 50.4%-49.6% Democratic win, which I really wished I published. The real results were 50.6%-49.4% and 51.04%-48.96% for Ossoff and Warnock, respectively. If I never do another model like this again, then I can walk off happy. Given that early voter turnout and data availability for 2022 are unknowns, I may never have to make another model like this again. We’ll see.
I’m proud of my strategic assessment in this race, and, unlike the 2020 Presidential Prediction, I laid out a realistic Republican path to victory.
Dislike: My only real dislike is some typos that I had in there. I’m a one-man shop; so, typos are going to happen. I probably have a couple former bosses who would have lost all confidence in the work because of the typos, but I don’t write for them anyways. 😊 And they would have ignored a good prediction.
What to Do Next Time
To be honest, I have no clue. I treat each election as a fresh prediction. Strategic assessments are the core of my predictions and that changes with every election. I’ll still rely on 538 and the Cook Political Report for good information. But my approach doesn’t lend itself well to midterms because there are hundreds of races. I’m also pessimistic about the early vote data availability. With coronavirus concerns subsiding (despite the ongoing pandemic), I’m not even sure if the early turnout will be there. Even though Democrats should push for that. The conventional wisdom has Democrats losing, and they’ll need every early and low propensity Democrat-leaning voter to turnout to give them a chance. Either way, I can make general outcome predictions without 2020-level early vote data.
I might focus on select Senate races to track moving forward. If I do, I’ll take what I learned about my Texas prediction, and get to know their races better before locking into a logical framework. I’ll also be tracking voter enthusiasm and turnout operations even closer. In many ways, weird political dynamics are persisting, and I’ll need to be flexible in my strategic assessments like I did in the presidential prediction. Otherwise, I have a lot to learn about the next election before deciding if I’ll make predictions. Once again, conventional wisdom relies on Democrats losing the House but potentially holding the Senate. But, in politics and in life, conventional wisdom can change very quickly. We’ll see if that happens here.
As for 2020, I definitely learned good lessons. But I leave cautiously confident as well. 3 races up and 3 races called correctly . . . Not a bad inning of pitching.
Sources:
[1] https://news.gallup.com/poll/276932/several-issues-tie-important-2020-election.aspx
[2] https://www.pewresearch.org/politics/2020/08/13/important-issues-in-the-2020-election/
[3] https://www.cnn.com/2020/10/30/politics/texas-2020-early-vote/index.html
[4] https://www.texastribune.org/2020/11/06/texas-trump-biden-counties-rural-suburban-city/
