Skip to content

The Economist not hedging the predictions

Andrew’s hedge that he’s predicting vote intent and not accounting for any voting irregularities either never made it to the editorial staff at The Economist or they chose to ignore it. Their headline still reports that they’re predicting the electoral college, not voter intent:

Predicting voter intent is a largely academic exercise in that all this hand-wringing over uncertainty in intent could be dominated by voting irregularities. For example, I applied for an absentee ballot in NYC on October 6. They have a tracker, so I can see it was approved on October 7, but never mailed. My inquiry to their help desk went unanswered. Now I have to decide if it’s worth voting in person in NYC amidst the pandemic. Mitzi applied a week before me at the end of September and received her absentee ballot two weeks later. Given the apparent gap in belief in Covid’s seriousness between the supporters of the two parties and the increased danger to the elderly from Covid, this feels like it could have a non-trivial effect on the election. At the end of September, erroneous ballots were shipped to Brooklyn. The absentee ballots themselves are apparently super confusing, but I wouldn’t know, since I didn’t get one.


  1. Andrew says:


    We’ve gone back and forth on the issue of how to handle potential voting irregularities in our forecast. On one hand, our model is based on fundamentals and polls, so it’s predicting some combination of vote intentions and turnout. On the other hand, you’re right that the goal is to predict the actual election.

    I’m still of two minds on this. As I wrote earlier today, I do think we should have longer-tailed errors to account for unpredictable events: in 2016 one unpredictable event is the possibility of large number of votes being counted (as discussed in commentary on a recent Supreme Court decision), and another unpredictable event is higher turnout, in part because people want to make sure their votes do count. There’s a logic to including these possibilities in the error term.

    As to the question of how to handle specific issue of ballots not being counted: I’m not sure. My Economist colleagues performed an estimate of the possible effects of not counting all the votes. I guess maybe I like to have this estimate separate from our vote forecast. Both questions are important: Which candidate is getting more votes in each state, and which candidate is declared the winner if there are legal manipulations.

    My plan is to just vote on election day as usual. Our polling place is never crowded anyway.

  2. Joshua says:

    Bob –

    > Given the apparent gap in belief in Covid’s seriousness between the supporters of the two parties and the increased danger to the elderly from Covid, this feels like it could have a non-trivial effect on the election.

    Perhaps at least part of the reason why it might feel like a non-trivial possibility to you is the emotional valence of what that possibility might mean to you?

    But what measure might you use to determine how trivial or non-trivial that impact might actually turn out to be? How could we even begin to make that determination? Have you seen attempts to do so?

    • Andrew says:


      Pollsters regularly ask respondents what are the most important issues to them. For example, I did a quick google and found this recent survey:

      Texas voters are casting their ballots with widely different opinions about the most important problems facing the country and the state of Texas, according to the latest University of Texas/Texas Tribune Poll.

      The most important issue facing the country right now is the coronavirus/COVID-19, chosen by 18% of registered voters, followed by political corruption/leadership (14%) and the economy (10%).

      What’s most important depends on who’s talking, however. Among Democrats, the most important issues facing the U.S. are coronavirus/COVID-19 (29%), political corruption/leadership (20%) and health care (11%). Republicans rank problems differently: moral decline (18%), the economy (13%) and political corruption/leadership (11%). . . .

      • Joshua says:

        Andrew –

        My guess is that breakdowns between pubz and demz about how they rank the issues is essentially a breakdown of whether they’re pubz or demz. The hierarchy of issues will obviously be influenced by what they think is more consistent with the candidates and party they support. It’s like what Dan Kahan says about opinion polling in climate change: asking opinions on that ranking isn’t really asking what you think, but who you are.

    • Joshua says:

      I’ll note that Andrew linked an anlysis. But it really only describes a theory of a potential relative disadvantage for demz. In the end it might be a non-trivial impact on the election but it could easily be more than overcome by an unusual level of turnout among demz relative to pubz. Or maybe the effect would be a good bit smaller than an unusual turnout among pubz eslrice to demz.

      So then how do you actually get away from Andrew’s “hedge?”. What is the actual alternative to assessing intent to vote and saying there’s not much you can recall do behind that?

      Uncertainty can suck. Especially when the uncertainty seems to have such importance.

    • Tom Passin says:

      “Risk” is often discussed as if it is nearly the same as probability, but actually it has another dimension: the impact. If the impact is low, fine, take a chance. If the impact is high (as in “death”, say), probably don’t take the chance. In the case of covid, an older person like me might rate the potential impact to be especially high.

      I voted early in person, but not the first few days it was possible – the lines were long. Hardly anyone was there, I was in and out in about 3 minutes, and I felt pretty safe.

    • Joshua says:

      I had a discussion with a friend today about which would feel worse: Trump wins because he clearly gets more votes, or Trump wins because Kavanaugh et al. stop the counting based on (what seems to us to be) a transparently partisan legal analysis.

      I felt the former would be more depressing because it would be such a sad statement on the country in which I live. My friend felt it would be the latter because it would be such a sad statement on the country in which he lives.

      • paul alper says:

        My feelings exactly.

        • Anoneuoid says:

          So let’s talk about where we are now. Just recently, there was a plan to kidnap Michigan Governor Gretchen Whitmer. The Center for Strategic and International Studies issued a report just a few days ago that white supremacists and other like-minded groups have committed a majority of the terrorist attacks in the U.S. this year. Earlier this month, the Department of Homeland Security warned that violent white supremacy was the most persistent and lethal threat in the homeland.

          A couple people arrested for this plot are on video saying trump is the enemy and they want all races to work together. Another was interviewed while attending a BLM protest back in June. Should take 10 minutes of research to figure out they are the opposite of racist pro-Trump right wingers like this implies.

          Looks like NPR is spreading dis/misinformation, and who knows what else they got wrong if these basic facts are wrong. This has been my experience with NPR in the past which is why I stopped listening about 5 years ago.

          • somebody says:

            > Should take 10 minutes of research to figure out they are the opposite of racist pro-Trump right wingers like this implies.

            Maybe you should have done more than 10 minutes of research. Among the indicted suspects include people who fly Confederate flags at their house, 3 percenters, anarchists, BLM supporters, QAnon conspiracy theorists, and at least a few Trump supporters. The fact is that not everything in the world can be easily projected onto a left-right political axis. The only real binding thread with these guys is that they dislike government overreach, really like the second amendment, and are a little unhinged.

            • Anoneuoid says:

              > Daniel Joseph Harris attended the rally out of concern for police violence.  “It is a shame what happened with George Floyd and instances where law enforcement officers murder an unarmed man/woman who isn’t resisting arrest, was complying with the orders is wrong and need to be stopped. You look at photos and videos of news teams and protesters being beaten by riot police when they are there peacefully, you see people losing their eyes because of an officer shoots them with a “non-lethal” round like pepper balls, or rubber bullets,” Harris said .  “I’ve gone through that sort of training and I can promise you weapons like that they can turn into a lethal round.“We went to the BLM protest yesterday in Lake Orion to show our support that everyone’s voice should be heard, no matter the color on your skin. Protesting is important to me because it gives us all a voice to be heard,” he said.




              Sounds like some real white supremacists there… Its a fake narrative, give it up.

            • I’m from Michigan originally and am pretty shocked by the level of gun ownership there across the political spectrum. I went to my uncle’s funeral last year in the Detroit suburbs, and the topic came up at the table of all the violence in the country. Turned out every other man at the very large banquet table other than me and my dad was toting a gun, either on their person or in their car (it’s not just men—lots of women I know there are into guns, too). They thought that was the obvious solution to gun violence—more guns. I’m regularly shocked when I get to the outer suburbs of Detorit and walk into a diner and see dudes with pistols tucked into their belt like a gunfight’s going to break out over pancakes.

              • Martha (Smith) says:

                Makes me glad that a lot of my relatives have moved out of Michigan.

              • rm bloom says:

                They’re tired of taking guff from everyone (especially their wives who bring in half or more the income nowadays) and so their manhood is on display. It’s hard not being the boss anymore. But any twit can be the boss with a pistol.

              • Martha (Smith) says:

                rm bloom said,
                “They’re tired of taking guff from everyone (especially their wives who bring in half or more the income nowadays) … It’s hard not being the boss anymore. ”

                I grew up in Michigan (Detroit proper) and saw the phenomenon of wives bringing in half or more of the income start in the early 1960’s. The company my dad worked for as an engineer moved to a state with lower taxes for corporations. He could have moved with the company, but that would mean disrupting my mom’s work on a graduate degree, as well as leaving elderly parents behind. Fortunately, he was both flexible and enterprising — he took some part time jobs, and used the fund for us kids’ college educations to buy an apartment building and put some sweat equity into it to make it a better investment than a bank account. And when my mother finished her graduate degree, she got a good job offer in Florida — and since the folks there said that my dad would easily be able to get a good job there, they moved — and indeed, my dad did quickly find a good (enjoyable as well as good paying) job. They were both very happy with the move.

              • rm bloom says:

                And while I’m on a roll with my amateurish social-psychologizing, I cannot resist offering up the secret behind their angry resistance to the stupid masks: the mouth and teeth are no longer visible — interferes with the favored habitus: aggression.

              • Martha (Smith) says:

                rm bloom said,
                “And while I’m on a roll with my amateurish social-psychologizing, I cannot resist offering up the secret behind their angry resistance to the stupid masks: the mouth and teeth are no longer visible — interferes with the favored habitus: aggression.”

                Gave me a good laugh. Thanks!

            • Joshua says:

              No one in the interview called the kidnappers white supremacists. The entire interview is about the movement more generally and about the Path Keepers specifically.

              > Atlantic writer Mike Giglio profiles the Oath Keepers, a pro-Trump militia group, in a new article.

              > My guest Mike Giglio writes about one such militant group, the Oath Keepers

              And again:

              > So let’s talk about where we are now. Just Recently, there was a plan to kidnap Michigan Governor Gretchen Whitmer. The Center for Strategic and International Studies issued a report just a few days ago that white supremacists and other like-minded groups have committed a majority of the terrorist attacks in the U.S. this year. Earlier this month, the Department of Homeland Security warned that violent white supremacy was the most persistent and lethal threat in the homeland.

              You shouldn’t hold others accountable for your misperceptions.

              • Anoneuoid says:

                I already quoted the part about the kidnapping and implied (but false) connection to white supremacists?

                They were the opposite of white supremacists.

              • somebody says:

                > They were the opposite of white supremacists.

                I know we’ve been over this already, but it’s just not true. Multiple suspects are affiliated with the three percenters which are explicitly right wing and whose members have committed white supremacist terror attacks. Others identify themselves with “boogaloo” which technically isn’t a white supremacist movement but does advocate for race war accelerationism. Two co-founders fly a confederate flag on their lawn. “Saying ‘the races shouldn’t fight’ but flying a confederate flag doesn’t constitute “the opposite of a white supremacist” unless you consider anyone who isn’t a white supremacist” “the opposite” of a white supremacist.

              • Anoneuoid says:

                As was pointed out already, I had already posted a video of Pete Musico that in his own words directly contradicted your claim he was a white supremacist. Keep repeating falsehoods if you want.

              • Joshua says:

                > I already quoted the part about the kidnapping and implied (but false) connection to white supremacists?

                You assumed the implication. It wasn’t made. It’s pretty harsh to classify something as mis/disinformation because you assumed an implication that wasn’t there.

                The topic of the interview was the Oath Keepers. She was listing information related to the trend and recent events leading up to the more specific focus of the interview. Nothing stated was incorrect (unlike what you said). And further on in the interview there was information given that most likely actually corrects misperceptions about whether all of these groups are white supremacists.

                Not to mention that I’ve seen this point raised many times in blogs. I mean more accurate information is always good – but I find this point of focus about whether reactionary, violent, extremist, militia groups are also white supremacists or just reactionary, violent, extremist, militia groups as opposed to other groups that are reactionary, violent, extremist, AND white supremacist is a bit misplaced.

                It tends to look to me a bit like deflection from the more key elements.

              • somebody says:

                I never once said Pete Musico was a white supremacist. I said he was a vocal Trump supporter (in contradiction to your claim that the kidnappers were the opposite of Trump supporters). I also stated that he flew a Confederate flag and shared white supremacist content on his social media pages, both of which are simply facts, photographs and archives, along with some other examples of racist or racist-adjacent behavior by suspects in on the plot.

                You’re acting like this is a debate about whether or not the kidnapping group is a white supremacist group, and like I’m saying it is. I am not. I’m saying the premise of this argument makes no sense. In trying to debunk the insinuation that the kidnapping group was a Trump supporting white supremacists militia you insinuated that they were an anti-Trump, BLM loving militia, which is at least as false. I am just debunking that claim.

                Maybe read more carefully in general, since you also seem to take the transcript of an interview as a news article.

              • Anoneuoid says:

                Yes, I always list unconnected information like this:

                Joshua posts bad things on gelmans blog. It has been reported that white supremacists and similar are big problem online.

                The intent to deceive is obvious.

              • Joshua says:


                You said it was stated. It wasn’t stated.

                You said that the interviewer knowingly spread mis/disinformation.

                First, you have no way of knowing whether that would be true unless you’re a mind-reader.

                But further, the evidence is the opposite, because the interviewer asked for clarification as to whether the topic of the interview, the Oath Keepers, were white supremacists. Obviously, the interviewers objective was to inform the audience, accurately, about who they are.

                But I am still curious. I have seen rightwingers focusing in blogs on the incorrect perception among some people about about some of these reactionary, violent, extremist, militia groups and whether they are white supremacists as opposed to the other reactionary, violent, extremist,militia groups that aren’t also white supremacists. Now not assuming that you are a rightwinger – I’m curious as to why in a long interview about a group of reactionary, extremist, violent, militia where it is clarified that they aren’t white supremacists, you focus in on an ambiguois statement that might imply another reactionary, violent, extremist, militia group that actually plotted to kidnap a governor is white supremacist. And then wrongly classify that ambiguous statement as mis/disinformation based on mind-reading abilities you couldn’t possibly have?

                And I hope to get an answer but before Andrew gets pissed off I’ll leave it there unless you do answer.

              • Joshua says:

                OK, I lied. Just one more point. Your contention was that the interviewer’s goal was to spread dis/misinformation, indirectly and by implication, about a particular violent extremist group while simultaneously making it a point to clarify information about another violent extremist group that was actually the subject the interview?

                Yeah ok. That makes a ton of sense.

              • Anoneuoid says:

                I agree that we should leave it at that. This really isnt place for it.

              • Don't feed the Anoneuoid troll says:

                Wasn’t it already established in this blog that “Anoneuoid” often acts as a ill-intentioned troll and shouldn’t be “fed”?

          • Joshua says:

            Anoneuoid –

            Actually, there is nothing inaccurate about this statement:

            So let’s talk about where we are now. Just Recently, there was a plan to kidnap Michigan Governor Gretchen Whitmer. The Center for Strategic and International Studies issued a report just a few days ago that white supremacists and other like-minded groups have committed a majority of the terrorist attacks in the U.S. this year. Earlier this month, the Department of Homeland Security warned that violent white supremacy was the most persistent and lethal threat in the homeland.

            Granted, the implication might be that the kidnap plotters were white supremacists, but it isn’t explicitly stated and if you bothered to listen to the interview the question about white supremacist was asked and answered by the reporter. So actually…

            > NPR does “say” it, actually the wording used implies they were white supremecists indicating it was not a mistake and they knew it was false.

            Is just wrong. Listen to the interview or just read the transcript. The interviewer didn’t “know w it was false” and explicitly asked for clarification and the question was clarified.

          • Joshua says:


            GROSS: Rhodes doesn’t want the Oath Keepers to be called a militia or to have its members be called white nationalists. He doesn’t want to be labeled as racist. In reality, is the group white nationalist and racist? Are they a militia?

            GIGLIO: I had a number of conversations with Stewart and with members of the Oath Keepers about the white nationalism label, and they didn’t just reject the label, they criticized white nationalism. They said that they’re part of the problem. They like to say that white supremacists and leftists are all part of the same basket. They’re all part of the problem. And so the way that they see themselves at least is that they are very much not white nationalists. Rhodes actually – and I put this in the article – he considers it to be a smear to be called a white nationalist. It’s a very difficult issue to say, like, every member of the group and what they believe. And, you know, the vice president of the Oath Keepers, for example, is African American. And they have from the beginning put a disavowal of racism on their site and banned racism officially. And the introductory video, even on their website is of a Black member. Like, they really want to push back against this. I end up deciding that the best way to position them is just Trumpist in their views about race. They are aligned with the president when it comes to Black Lives Matter. So to them, at least the leadership level, Black Lives Matter is a Marxist group, which I would note is a foreign enemy. You know, so we’re talking about enemies, foreign and domestic. This is a foreign enemy on domestic soil if they’re calling the Marxist. But President Trump says that as well.

      • Dzhaughn says:

        Perhaps it would be saddest to have it be the former and believed to be the latter.

        (Caveat: The former = “winning the electoral college.” Don’t give me any of this national popular vote second guessing. Should we put asterisk next to teams that have won the World Series but scored fewer total runs?)

    • It’s definitely emotional as I live in the U.S. and on Earth. The two factors I mention would pull in the opposite direction on the vote.

      I’m old enough to be worried about Covid and considering not voting for the first time in my life. But I’m not so old I’ve forgotten the 2000 election. Or the 2016 polls (which could’ve been right—1/10 isn’t that extreme, despite everyone seeming to think 90% or 97% is a sure thing—I wouldn’t fly in a plane with a 10% or even a 3% chance of crashing).

      I pretty much only follow polling to the extent Andrew blogs about it, so I haven’t seen any attempts to adjust for possible voting irregularities, the effect of Covid on turnout, etc. I have no idea how you could do that without more polling and some very strong assumptions.

      • jim says:

        As long as there aren’t people crowding into rooms for extended periods (>20min), you’ll probably be OK. it’s worth checking it out at any rate.

        I’m sorry to hear about your woes with mail in voting. We’ve been doing it for quite a while in WA and it works great. We also have drop boxes everywhere.

        • dhogaza says:

          And Oregon’s first vote-by-mail law was passed 39 years ago (optional for county level elections, with at least one polling place open in any county that adopted it).

          Statewide vote-by-mail with election day drop-offs was in 1996. That’s awhile ago :) Works great. Oregon Republicans have stopped opposing it, polls show high support among rank-and-file party members. Ask an Oregon native and they’re likely to respond with something like “there are other ways to vote???”

          When I was younger I could’ve recited dates and all off the top of my head. It’s been around so long in Oregon (where I lived until 6 years ago) that I had to look it up in Wikipedia to get exact dates.

          Opponents to making it statewide for all elections cited – and I’m sure this will make you gasp with surprise – concerns about voter fraud. Which have never materialized in the 39 years since Oregon began experimenting with vote-by-mail.

      • rho says:

        What does 10% chance of X winning an election mean really? If we run the election 1000 times, we will find that Trump will win about 100 times?!!

        I wish the model output are framed properly. Journalists may find it challenging to understand, but I bet a lot of readers will get the message.

  3. paul alper says:

    Bob Carpenter nailed it:

    “Andrew’s hedge that he’s predicting vote intent and not accounting for any voting irregularities either never made it to the editorial staff at The Economist or they chose to ignore it. Their headline still reports that they’re predicting the electoral college, not voter intent”

    Given the actions of the Supreme Court to suppress votes which arrive late due to postal problems and the right-wing militia ready to intimidate voters, the relationship between intent and outcome is less than solid. A complicated model that ignores the gorilla in the room is of limited use. From

    “A problem or difficult issue that is very obvious, but is ignored for the convenience or comfort of those involved.”

  4. Dalton says:

    Here in Washington State we all vote by mail. I live in a small town (pop 500) in a rural county (total population < 30K). I dropped my ballot off a dropbox I ride my bike to. So we have it better than most. Still I checked my ballot status and it was challenged because the signature didn't match. I called the county. They check every signature "by hand" or eye. A human decides whether or not you're signature matches the one on file. There's a fairly cheap application of Bayes rule here, but I am willing to bet that while the probability of rejecting a ballot is low (around 1%), the percentage of ballots stolen and the signature forged is way less than 1%. They do have to notify you if your ballot was challenged, but I am miffed. Our state has way more experience with this then most states, but we still have a 1% rejection rate which is still a lot of potentially disenfranchised voters.

    • Martha (Smith) says:

      I can understand how you are miffed!

      I also voted ballot-by-mail, but here was an intermediate step: Here, you don’t drop the ballot in the box, but drive (or walk) up to a person who takes your ballot (in its signed envelope), gets your signature on a registry, and checks the signature against your drivers license or other ID.

      I have been wondering what the chances of a signature being challenged are (my signature sometimes seems all over the place — and definitely sloppy), but am hoping that between the signature on the registry and the one on my ballot, that it will not be challenged.

    • aosmith says:

      Do they give you a chance to correct it? In OR we have 14 days to go in and correct things if the signature is challenged (I have never had to do this so don’t know how the whole process works in real life). I’m sure folks who don’t have time to go in are still disenfranchised but I certainly see the process as an honest attempt to allow people to vote while still trying at fail-safes.

  5. fogpine says:

    Predicting voter intent is a largely academic exercise in that all this hand-wringing over uncertainty in intent could be dominated by voting irregularities.

    Yes, I think you’re right Bob. All my, Andrew’s, and Elliotts comments about miscalibration, negative between-state correlations, The Economist vs 538, etc. are far less important than the issue of voting irregularities. By voting irregularities, I mean the COVID-19 type that you mention, but also the chances that Trump refuses to accept election results and manages to stay president.

    The prediction markets do try to account for COVID-19 effects and scenarios where a candidate becomes president non-democratically. So I’m curious if differences between predictions from the markets, The Economist, and 538 are larger for states where Trump’s party controls the legislature and/or relevant courts.

  6. Adam says:


    If you’re up in Harlem, I voted today (Wadleigh HS) and the process was very smooth. Everyone was masked up, they had stickers on the floor for distancing, and everyone was very professional.

    I was in and out in about 10-15 minutes.

    • I live in the Village. I went today at 2pm. There was a line down the street that turned the corner, but nothing like 2016 at the usual voting place.

      Before we got to the door, there was a TV crew with some group passing out pizza, so everyone bunched up and took their masks off to eat.

      When I got inside, it wasn’t as crowded as voting usually is (like the subway at rush hour), but neither was it socially distanced. They kept telling me to move closer to the person in front of me in line. I asked about social distancing and the pollworkers said they were doing the best they could with the space they had, which wasn’t enough to maintain social distancing with the number of people they were letting in at a time.

  7. David says:

    What happened to you is odd. I also live in NYC. I also requested my ballot in late September. The BOE website claimed it was mailed on the 29th-but by October 10th or so I was suspicious it would never come. I called the hotline 2x (35 minute wait each time)-on the second call by asking questions I found out they mailed it to the wrong address. I simply asked what to do-the woman said go to the BOE (200 Varick St.) and show them your confirmation number and they will give you a ballot. There was almost no one there, they were very helpful, and I filled it out and left it with them. Took 15 minutes and today I checked that it was accepted online.

    I doubt these things are all that common (my brother and sister in NYC did not have any issue)-but the BOE was helpful and there was not any difficulty fixing the issue.

  8. Rahul says:

    Why do we obsess so much about predicting election results? Especially in the US.

    Of course, elections matter. But does predicting them make things better on a societal level. I.e. what’s the societal value of devoting so many resources to election prediction? Or is it a case of I predict because I can?

    I can see how predicting other things can be of value: eg rain, earthquakes, heart attacks etc.

    • rm bloom says:

      I was puzzling over this very thing too. I also asked ‘to what extent do these predictions themselves influence the outcome?’

      Much practical prediction work goes on the premise that it helps us be better prepared.

      So Newton’s laws (or Kepler really) taught us how to predict the conjunction of the planets so that we know better where to look.

      As a side-benefit it advanced tremendously the design of armaments like cannons to hit their mark.

      Statistical models of rare events help us to build bridges that don’t come down every 100 years — that’s what I’ve heard from Gumbel and the engineers. Maybe too this helps with earthquakes.

      Also samples on the factory line taken right obviate the need to test every damn part, at least under stable conditions; and then you get to see the departure from stability, in the uptick in prediction error.

      What better better position does prediction of elections put us in? Maybe it’s supposed to help get better sleep on election eve; or maybe, sleep worse?

      Maybe these predictions aren’t for us, but are really commissioned by campaigners, who seek to know better whether or not their tactics work?

    • jim says:

      Prediction models probably emerged originally for candidate strategic use and that’s where they make sense.

      But they’re also just popular with hipsters because it’s the Age of Data. Just machines to make big decisions, programmed by fellas with compassion and vision…

      Most people aren’t sitting around wigging out about this stuff.

      • Rahul says:

        I just think it’s a collosal waste of resources. To the extent of strategic planning by the candidates I can see the utility.

        But for academics to spend so much time on it seems a utter waste.

        • Andrew says:


          There’s an academic value of forecasts in that they help us understand voting. A key component of our 1993 article, “Why are American Presidential election campaign polls so variable when votes are so predictable?”, was that elections are indeed predictable.

          How much time to spend on it is another question. I find it valuable to spend time on almost any problem—God is in every leaf of every tree. An advantage of working on a popular topic such as sports or politics is that you can get many different outside perspectives on your work. I’ve learned a lot about forecasting and modeling in general by focusing in depth on the problems of our election forecasting model.

    • rm bloom says:

      I have discovered or re-discovered what I think is the reason we seize upon these forecasts; it is a variety of “whistling past the graveyard”. But it is like that story “Appointment in Samarra”. Here is Maugham retelling it in 1933. The narrator is Death itself:

      There was a merchant in Bagdad who sent his servant to market to buy provisions and in a little while the servant came back, white and trembling, and said, ‘Master, just now when I was in the marketplace I was jostled by a woman in the crowd and when I turned I saw it was Death that jostled me. She looked at me and made a threatening gesture, now, lend me your horse, and I will ride away from this city and avoid my fate. I will go to Samarra and there Death will not find me.’ The merchant lent him his horse, and the servant mounted it, and he dug his spurs in its flanks and as fast as the horse could gallop he went. Then the merchant went down to the marketplace and he saw me standing in the crowd and he came to me and said, ‘Why did you make a threating getsture to my servant when you saw him this morning?’ ‘That was not a threatening gesture,’ I said, ‘it was only a start of surprise. I was astonished to see him in Bagdad, for I had an appointment with him tonight in Samarra.’

    • Baseball doesn’t matter, either, but there’s a huge industry around it. I wonder how it compares in size to the polling industry in both statisticians and feet on the ground. Of course, there there’s a good reason to predict baseball events—you want to win games or make book.

      My guess is that as long as people want polls, someone will produce them. But that’s just the supply side—why do people want to consume these? I think it’s because they want some degree of certainty—I’m surprised at how many people jumped on 2016’s predictions as wrong; saying an event has a 10% (or even 3%) probability doesn’t seem that low to me, but at least the latter we could reject at p = 0.03 :-)

      @rm bloom: Don’t forget the tides! Ships and cannons were very important to Britain back when they ruled the waves. Unsurprisingly, Newton’s tide theory was later refined by Laplace. Also, lots of factories are using the “internet of things” to instrument every aspect of their production line—Alp Kucukelbir, who built Stan’s variational approximation, runs a company doing this. My guess is that waiting for the finished product to go wrong can be much more costly.

      @jim: Jill Lepore wrote a book on Simulmatics Corp., the analytics company that Kennedy used in the 1960 election. There’s a short version in the New Yorker. I’m sort of surprised Andrew hasn’t blogged about it already.

    • Mendel says:

      Humans naturally hate uncertainty. They get anxious. (The aim of terrorists and torturers is to produce uncertainty.)
      To combat this anxiety in the face of an uncertain future, people have consulted fortune tellers and astrologers; and that is why they pay statisticians.

  9. Anon says:

    Shouldn’t a higher priority be put on adding this to the modeling of uncertainty? It’s not some small thing but probably the most important part of the election.

  10. Marc says:

    Seems like another case where the headline writers make a much bigger claim for the research than a careful read of the paper shows. If only someone who knew and cared about these things had the ear of the headline writers ….

    More seriously, I think that by now there is much more uncertainty in how the vote occurs than in the voter intention measured by polls and assessed by the Economist model. We have: a) how turnout may be affected by the combination of COVID and increased access to early voting (these go in opposite directions, making any prediction more difficult), b) any chaos — intentional or not — on the day of election, and c) controversy over how votes are counted.

    I don’t have any great idea of how to quantify these, but I think the best summary of the state of the election is that voter intention is solidly in favor of Biden in enough states to win the electoral college, but that uncertainty of how the vote actually happens and is counted means that there is a non-negligible chance that Trump could win.

    Thus the most important messages are : VOTE and do whatever you can to ensure that polls are accessible and ballots are fairly counted.

  11. KL says:

    Election irregularities seems like banter about the error term. Every election has an error term.

    The last election was unprecedented, with a reality star businessman running in the age of Twitter.
    Previously, we had the first black president.
    Before that, we had 9/11.
    Before that, we had the vice president from an administration where the president was impeached.

    Voter intentions are not observable. The popular vote share is observable, but everyone cares about the next president.

    • Andrew says:


      You say, “Voter intentions are not observable.” That’s not correct. Voter intentions are observable. They are observable by simply counting all the votes. When I say “voter intentions,” I’m speaking of the intentions of the people who want to vote. One of the controversies of the current election is that there is the possibility of people who want to vote but not having their votes counted. That is not included in our model.

      • fogpine says:

        One of the controversies of the current election is that there is the possibility of people who want to vote but not having their votes counted. That is not included in our model.

        I can’t find where you say this. Bob Carpenter points out that your hedge/disclaimer isn’t mentioned in the Economist web page’s headlines. I also checked the rest of the Economist’s national presidential forecast webpage, state presidential election webpages, senate forecast webpages, house forecast webpages, and “how it works” explainer webpages. I don’t think your hedge is mentioned anywhere in the more than 100 webpages on the Economist website.

        On the Economist website, your forecast is explained using sentences like, “The bars below represent the predicted likelihood of every plausible electoral-vote outcome.” To me, “every plausible” indicates no major hedges/limits to the applicability of predictions.

        So I think that the large majority of people trusting your 4% Trump win probability don’t understand the hedge is present.

        This was less important when the model was predicting a higher Trump win probability. However, the model now gives Trump a 4% win probability, Trump is openly suggesting many ballots should not be counted, and COVID-19 rates are rising! In this context, the hedges/disclaimers on your predictions matter much more.

      • Kaiser says:


        In addition, polls are a good measure of voter intentions as we ask people who they intend to vote for. One can complain that people lie, don’t respond, etc. but it is directly measurable.

        It makes perfect sense to separate out voter intention from counting irregularities.

      • KL says:

        Thanks for your response. Perhaps some states will continue to count while appealing a court reporting deadline, and courts will reject the final totals. Otherwise, I don’t see how we would know that your prediction was correct.

        On a different topic, your model has higher interstate correlations, and Nate Silver’s model has fatter tails. It is curious that these modelling differences offset to produce similar predictions. Do you have any sensitivity diagnostics on this? For example, how would predictions change if your model used Nate Silver’s correlations, or if your model used fat tails?

        • Kaiser says:

          The model differences are more complex than that, though. The FiveThirtyEight model has relatively consistent interstate correlations regardless of state and consistently wider ranges than the Economist model. That implies they are not using state-specific information on interstate correlations while for example, in the Economist model, all the DC interstate correlations are bunched up at a low value, which implies that Andrew and associates believe that the DC vote share is mostly independent of what happens in other states (which makes sense to me). See the boxplots in my post here.

        • Andrew says:


          Regarding “observable”: Not everything that is observable is always observed. In statistics we distinguish between these two concepts.

          Regarding the correlation and state uncertainties: It’s no coincidence. The Fivethirtyeight team and the Economist team (i.e., us) had similar aims when it came to the national forecast. The Fivethirtyeight team included some low and negative between-state correlations (by accident, I think) but the result was that they needed super-wide uncertainty intervals at the state level to get their desired uncertainty for the national popular vote and electoral college. That’s how they ended up with things like Biden having a 6% chance of winning South Dakota.

          • KL says:

            “The Fivethirtyeight team … needed super-wide intervals at the state level to get their desired uncertainty.”

            That is exactly my concern. If they already knew the answer they wanted, then why did they need a model? It sounds like they backfitted a model to justify the desired answer! It reminds me of Feynman’s “cargo cult science”.

            Fat tails seem unimportant, because the election will probably be decided by narrow margins in a few battleground states. The models must be very sensitive to the correlation assumptions. I think those correlations are hard to measure. It would be good to generate predictions for the realized covariances across 52*51/2 = 1,326 state vote shares. If those predictions are bad, then a good electoral prediction is just a coincidence of error cancellations, and it might be better to make future predictions using a simpler and more robust framework.

            • Andrew says:


              Even if you know where you want your inferences to be, the model is still useful in helping you incorporate new data. Recall that a big motivation for the Fivethirtyeight forecast was that simple poll aggregation has problems because polls are of uneven quality. We take it a step further and integrate state and national polls. Ultimately the inference is only as good as your assumptions, but a model does allow you to fold in new data in a reasonable way.

  12. Christian Hennig says:

    One issue that is nicely illustrated here is that the statisticians can be as skeptical and careful as they want about their models, the general public will still take them as “assumed true”, i.e., as what the statistician/scientist really believes to go on in reality… as long as they at least don’t take it as the truth itself.

  13. Kaiser says:

    I’m going to try to express this thought in my head as clearly as I can. I’m a bit perplexed by it so… in building these models, the posterior distribution is supposed to capture the general picture, with “black swan” type events in the tails. Now, if Andrew’s speculation were true, FiveThirtyEight has info saying the wind is blowing in the direction of the black swan. So they try to nudge the distribution to show higher likelihood of a black swan event (make the tails thicker). Alternatively, the modeler can stick to the same model, arguing that the black swan event is already in the tail. But something doesn’t sit right. Whether or not there is wind direction about the black swan, we would then stick to the same predictions. What am I missing? It seems like FiveThirtyEight is on to something but maybe their solution isn’t the right one.

    • Andrew says:


      All events are possible but I think some of the conditional probabilities in the Fivethirtyeight prediction don’t make sense, also I think some of the state prediction intervals are too wide.

      • Kaiser says:

        I agree that their uncertainty bands are hard to explain, and I think the back and forth will improve the model.

        Let’s say FiveThirtyEight wants to express the viewpoint that this year may be the year where the unexpected happens. I’m wondering how our modeling apparatus deals with that.

  14. Mitzi says:

    just to let y’all know that this afternoon Bob and I walked down to the NYC early voting site near us and at 2pm in the afternoon the line was only about a 10 minute wait. and the Williamsburg Pizza guys were there handing out individually boxed slices of democracy pizza – although as Bob points out, questionable way to encourage mask wearing. thanks to all the NYC BOE volunteers for making this possible.

    • confused says:

      What kind of toppings does democracy pizza have?

      • Corn, Ethanol, and Pork Bellies

        • Martha (Smith) says:

          Sounds to me like a recipe for a belly ache.
          (Hmm — I didn’t intend the wordplay, but it fits!)

        • Martha (Smith) says:

          Also, this reminds me of the old card game Pit ( –as I recall, the version we had when I was a kid just had wheat, corn, barley, oats and rye as commodities. ). My grandmother loved it. (I think that might have had something to do with the fact that she was one of something like 13 children.)

        • jim says:

          Actually if you look at the “American Food” section in a French grocery store, it’s Lucky Charms, chips, pretzels and Skippy.

          Stream of conciousness:
          Skippy makes me think of Thai food with peanut sauce. Peanut sauce is awesome with spicy food. I love pizza with jalps and pineapple. What about peanut sauce, jalp and pineapple pizza!!! YUM!!!

        • Anonymous says:

          I think the point is, in Democracy Pizza, you just get what the senators from unpopulated ludicrously overrepresented states decide what you can have so their donors will make a bundle

          • jim says:

            Yes, we wouldn’t want the kind of democracy where California doesn’t call all the shots, would we?

            • confused says:

              CA is only 40 million out of 330 million… not enough to “call all the shots”, though it’s the largest state by population.

              Everyone talks about CA and NY, but they are really disproportionate culturally (due to media centers) not so much population wise. TX and FL are between CA and NY in population.

          • jim says:

            “senators from unpopulated ludicrously overrepresented states…”

            Some actual facts about representation in the US Senate:

            Republicans represent 48.1% of population and 66% of land area excluding DC.

            Democrats represent 51.5% of population and barely a third (33.4%) of land area

            There are two independent senators, in Maine and Vermont, who represent 0.3% of pop and 0.6% of land.

            Republicans represent two of the three most populated states: Texas and Florida.

            Republicans and Democrats each represent two of the four most populated states.

            Republicans and Democrats each represent four of the ten most populated states, with two of the ten having split representation.

            Non-Republicans represent five of the eleven least populated states: Hawaii, New Hampshire, Rhode Island, Delaware and Vermont.

            Democrats represent the last eight states on the list in terms of land area:  Maryland, Hawaii, Massachusetts,  Vermont,  New Hampshire,  New Jersey,  Connecticut,  Delaware and Rhode Island, the total land area of which is less than the state of Washington.

            Overall, population-wise, the Senate is pretty well distributed across the population.

            • Chris Wilson says:

              “Overall, population-wise, the Senate is pretty well distributed across the population.”

            • confused says:

              The population split is more extreme than the *by party* split since the D’s do have some low population states (HI, NH, etc.)

            • Joshua says:

              > Some actual facts about representation in the US Senate:

              The imbalance is seen in that a minority of the population controls a majority of the Senate. Obviously, it translates into white, rural voters having disproportionate power

              This leads to unequal representation. For example, Kavanaugh was confirmed by senators representing 44% of the population.

              In 2018, Democratic Senate candidates got 8% more votes but lost two seats.

              • confused says:


                I mean the Senate was never intended to be proportional — it was always meant to be “by state” not “by population”, and indeed originally Senators were selected by state legislatures rather than by popular vote in their states — but I think it is significantly more disproportionate now due to the vast growth of urban areas in some states in the last 60-70 years.

                States were always admitted with small populations, but early on they grew pretty quickly — but places like Wyoming and the Dakotas haven’t grown much in a century.

                However, *who* the imbalance gives power to is somewhat of a historical quirk. If Puerto Rico, DC, Guam, and the US Virgin Islands became states, it might look different…

            • Anonymous says:

              laughable… Here are the states by population


              The top 8 states have 16% of the senators and roughly 50% of the population

              the bottom 20 states have 40% of the senate and roughly 10% of the population

      • Mitzi says:

        When I said “democracy pizza” I was thinking of the Australian food “democracy sausage” which you get at a “democracy sausage sizzle” – cf the wikipedia:

        “Democracy sausage” is the colloquial name for a sausage wrapped in a slice of bread, bought from a sausage sizzle operated as a fundraiser at Australian polling places on election day, often in aid of the institutions that house the polling place. In 2016, the BBC reported that just under one-third of the 1,992 polling booths across Australia had a sausage stand by the count of the Election Sausage Sizzles site. The tradition goes back to cake sales in the early 20th century and is considered a patriotic duty of all Australians.

Leave a Reply