Answering some questions about election forecasting

Just in time for Halloween!

My colleagues and I have published a couple of articles about forecasts of the upcoming election:

Why forecast an election that’s too close to call? (in Nature)

Grappling with uncertainty in forecasting the 2024 U.S. presidential election (in Harvard Data Science Review, with Ben Goodrich and Geonhee Han)

And a talk:

Tough Choices in Election Forecasting: All the Things That Can Go Wrong (at the Washington Statistical Society)

And a zillion blog posts:

A question for Nate Cohn at the New York Times regarding a claim about adjusting polls using recalled past vote

Prediction markets and the need for “dumb money” as well as “smart money”

“Things are Getting So Politically Polarized We Can’t Measure How Politically Polarized Things are Getting”

“Pitfalls of Demographic Forecasts of US Elections”

Wendy Brown: “Just as nothing is more corrosive to serious intellectual work than being governed by a political programme (whether that of states, corporations, or a revolutionary movement), nothing is more inapt to a political campaign than the unending reflexivity, critique and self-correction required of scholarly inquiry.”

It’s martingale time, baby! How to evaluate probabilistic forecasts before the event happens? Rajiv Sethi has an idea. (Hint: it involves time series.)

Nonsampling error and the anthropic principle in statistics

Instability of win probability in election forecasts (with a little bit of R)

Election prediction markets: What happens next?

Why are we making probabilistic election forecasts? (and why don’t we put so much effort into them?)

What’s gonna happen between now and November 5?

Polling averages and political forecasts and what do you really think is gonna happen in November?

The election is coming: What forecasts should we trust?

etc etc.

And the occasional newspaper interview:

"Normale Menschen nehmen nicht an Umfragen teil" (Die Zeit)

Also sometimes they just send me questions. Here are some examples.

A journalist writes:

I am working on an article on why Donald Trump is the favorite in the presidential race according to the betting odds, but not the clear favorite in the opinion polls.

I will be grateful if you can help me with the following questions (even if you can answer just some of them):

1. Why is there a difference between opinion polls and betting odds? 

2. How exactly are betting odds on political events formed? Especially by Polymarket?

3. What should we understand about the US political prediction market? Who are the main players? Where do Americans place the most bets on political events?

4. The Wall Street Journal, the New York Times, and other media reported that one player bet $28 million on Trump's victory. Can this be considered betting manipulation? How do such bets affect the odds?

5. I have a hypothesis that Trump is considered to have a lot of ‘silent voters’. So a gap of a few percent in the polls in swing states could mean that Trump has an advantage in real life. Something similar happened in the 2016 election. Could you comment on this hypothesis, please?

6. Elon Musk repeats that bets are more accurate than opinion polls because money is at stake. Can you agree with this statement?

My reply:

1. On the plus side, bettors can use information not in the polls.  On the minus side, bettors can be influenced by their personal feelings.

2,3,4. Rajiv Sethi discusses these issues here:  rajivsethi.substack.com/p/forecasting-mechanisms-another-update

5. The issue is not that voters are silent.  The issue is that most people do not respond to polls, and it can be difficult for pollsters to adjust for differences between sample and population.

6. No.  People with money at stake can still lose money. Having money at stake gives a motivation for accuracy, but that doesn't mean you'll actually be accurate.

Another journalist asks:

I am reaching out to you for the following reason: I am writing an article about polls and different predictive models ahead of the upcoming U.S. presidential election on November 5th, and I would like to ask you a few questions to cite your analysis.

I would greatly appreciate it if you could take a moment to share your insights on this matter. 

Thank you very much in advance.

1. Firstly, I would like to ask you how the polls should be interpreted in this election cycle. So far, the various surveys show a very close race between Kamala Harris and Donald Trump, with differences between them within the margin of error in the key swing states. What are the main challenges that polling centers in the U.S. are facing when conducting opinion studies? Are there possible distortions—such as people not wanting to respond or a hidden vote for one of the candidates—that could affect the projections, or is the situation indeed one of an almost perfect tie in which anything could happen?

2. Secondly, I would like to ask you about predictive models in such a close election. Of course, I understand that the design of these “oracles” is becoming increasingly complex, technically speaking, and, in addition to polls, includes statistical adjustments, demographic factors, voting history, the state of the economy, and many other elements to try to calculate the probability of each candidate's victory. What have been the main changes or updates that had to be made to the latest design, considering the experiences from the 2016 election (when many “got it wrong” or, rather, the less likely option won) and the 2020 election? How challenging has it been to design a predictive model in such an unusual election, in which the incumbent president has withdrawn from reelection at the last minute, one of the candidates is a former president seeking to return to power, and there have even been unpredictable factors such as an assassination attempt?

3. Lastly, the million-dollar question: Based on your projections with the latest available data, is it possible to forecast who will win the U.S. presidential election? Who currently has the best prospects or the smoothest path to victory?

My reply:

1. The polls can be off.  There is non-sampling error, and we won't have a great sense of this until after the election occurs.  In addiition, public opinion can shift during the next two weeks.  Also, it is unknown exactly who will turn out to vote.

2. There are no oracles.  A forecast will use some sources of information.  At the Economist magazine, we are using a forecasting model similar to that of 2016, with some adjustments made to the structure of correlations of opinion shifts and state-level polling errors among states.  Regarding the change in candidates, we handled that just be shifting from Biden polls to Harris polls.  Enough polling data are available that there was no need to use Biden polls to estimate Harris's vote.

3. I've worked with the Economist magazine on their election forecasting model.  So I will answer that question by pointing you to The Economist's current forecast.

Another journalist asks:

I would like to ask you to answer a few questions about the US presidential election for my upcoming article. Do you think you could find the time to answer the following?

1. What role do disinformation and cyber threats play in this year's election? 

2. How can these things affect the election process and the overall security of candidates?

3. Are social networks definitely the most important place to build a campaign at this point?
4. How is it possible that people keep voting for Trump even after he has told countless proven lies?

My reply:

1, 2, 3:  These are important questions on which I have no expertise.

4:  Some voters like Trump, others believe the alternative from the Democrats is worse.

Another media request:

I am a freelance journalist based in Paris and a Sciences Po Paris graduate. I freelance for France 24, Reuters, the Green European Journal and other media.

I am reaching out to you with an interview request. I'm writing a piece about the importance of the vote abroad for swing states. Your expertise in statistics and political science will be a great match and will help me and the readers understand how the expat vote affects the big picture. 

We won't go into super big details, since it's for a European audience, but I want to know why the expat vote matters, some numbers on the difference it made during the previous elections and whether this year the stakes are higher. 

My reply:

Je ne suis pas expert sur les votes des américains expats. Je recommande Michael McDonald à l'Université de Florida.  Il est l'expert sur le taux de participation électorale.

And then this came in from a reporter from a U.S. national newspaper:

I was asked to work on a story about how Allan Lichtman and Nate Silver differ in terms of prognostication approaches and styles.

I’m looking to speak with an objective third party about their differences and you came to mind as the perfect person. Do you have 5-10 minutes to chat today about how people should think about Lichtman vs. Silver’s models and election forecasting in general?

My reply:

My quick answer is that reasonable forecasts make use of 4 pieces of information:  economic and political conditions compared to historical patterns, recent state-by-state election outcomes, state polls, and national polls.  Given this information, different methods will give very similar forecasts.  I recommend the forecast at the Economist magazine, which my colleagues and I helped on, and the forecast at Fivethirtyeight.com, which my colleague and I helped on a few years ago.

I also get some requests to go on TV. Fortunately, that hasn’t happened yet.