eagleclosed

The Washington Post

Shambala Preserve, Acton, California

Stringer

Homegrown tomatoes

Contributor

Surfside Tragedy Makes Condo Buying Challenging Nationwide

For The Washington Post

staff

Winter Hill offers single-family homes near Falls Church downtown

Maansi Srivastava for The Washington Post

A layer of ice coats the leaves of a Japanese maple tree after an ice storm in Toronto

Reuters

Daron Taylor

Could gun ownership help explain the partisan divide in hearing loss?

Science

Blueberry Picking in Full Swing at Butler's Orchard Farm

Staff Photographer

HOGP

Old town Lahaina used to be crisscrossed by canals and spring-fed ponds. But sugar plantations diverted the water and the wetlands were buried under baseball fields and parking lots. 

Pop-up kitten lounch for adoption in Georgetown

for The Washington Post

President Donald J. Trump

YEARINPHOTOS2018

DATA DEPARTMENT DOD RELIGION July 1, 2024 Promo image.

Couple In Their Backyard

International student workers who have come to this country to work for the summer are preparing to leave for the season, in Ocean City, MD.

Respond to prefers-color-scheme dark in app web views

Whether leaderboard ad sticky behavior is disabled

Follow section front redirects in Spectrum

Enable NLP source in PCS for load testing

Switch off the News Alerts Survey Widget when we have enough responses

Poll to determine if new LUF updates are available

Allows the winning A/B test variant to be turned on without Spectrum deployment

Show expanded byline information on articles

Show infinite scrolling sections in Assembler MultiView

Show option for NLP article summaries in Action Bar

Show NLP summaries collection recirc on article pages

A general feature flag for Recipe Features that aren't yet released

Show the Balance of Power component on the new Election subnav component

Feature flag for media team features that aren't yet released

Enable new Elections subnav on Site pages

Use new Orca video player for preroll plugin in Spectrum

If true, will use a spoofed user agent to scrape and summarize content from paywalled sources

flip back to old full page reload when new updates come in

Show the snow tracker (instead of the heat tracker) on Spectrum

Vice President <a href="https://www.washingtonpost.com/kamala-harris/" target="_blank" class="contextual_link">Kamala Harris</a> will win the 2024 presidential election. The tomatoes have spoken.

Over the past 17 straight election cycles, those fragile fruits have divined the fate of the republic. When a very specific measure of their prices falls, the Republican has won, starting with Dwight D. Eisenhower in 1956. When it rises, the Democrat always triumphs, as Joe Biden did in 2020.

This time around, the 12-month average of U.S. urban tomato prices rose 32 months before the election — predicting a clear victory for Harris.

Really! The math is legit. We did it ourselves. But “legit” in this case may not be a synonym for “useful.”

We dug up the tomato streak — along with millions of others like it — to evaluate the efficacy of what we’ve long suspected to be the least helpful sort of election forecasting: the magic indicator.

Department of Data Column Branding

We here at the Department of Data are dedicated to exploring the weird and wondrous power of the data that defines our world. <a href="https://washingtonpost.com/business/2022/06/29/dept-of-data/" target=_blank>Read more.</a>

One occupational hazard of writing a data column — even one that usually avoids the crowded “<a href="https://www.washingtonpost.com/people/lenny-bronner/">election-predicting wunderkind</a>” lane — is that certain folks ask for your take on tabloid stories or social media posts about this or that genius method that has predicted the last nine (or 11 or 13) elections. (For the sake of argument, let’s call these folks the “mothers-in-law.”)

As with the tomatoes, the math usually pencils out. These indicators can run up substantial streaks. But in almost every case, folks weren’t sitting around waiting for Ike’s election results in 1956, predicting that their newfangled tomato-price measurement would guarantee Adlai Stevenson’s stomping. Instead, these magic measures have emerged in recent years with the substantial help of hindsight.

That, of course, makes finding the magic indicator much easier. We know because we tried it. We fired up a massive data grinder and fed in more than 2,000 data series running back to at least 1960. We looked at everything: consumer spending on hotels, hours worked by loggers and miners, the price of hay and hayseeds. Then we tortured the numbers until they coughed up a prediction — either about the incumbent winning or losing, or about the Democrat/Republican winning or losing.

And guess what? We found almost 87,000 cases in which one of our convoluted variable transformations correctly picked at least 10 elections in a row. A whopping 500 of them hit at least 16 straight elections. And one variable — all hail the 37-month average commodity price for raw cotton — has successfully called 20 straight presidential contests, and counting! If we’d had unlimited patience and computing time, we could have found many more.

What does this mean for 2024? Together, 59 percent of our various indicators say the Democrat will win, pointing to a victory for Harris. But 74 percent say the incumbent party will lose, pointing to a victory for former president <a href="https://www.washingtonpost.com/donald-trump/" target="_blank" class="contextual_link">Donald Trump</a>. (In particular, measurements taken between three years and 18 months before the election, when inflation remained high, tend be about 20 percent in favor of the incumbent. Recent ones give the incumbent Democrats a somewhat better shot, climbing toward 40 percent.)

But should we take the streaks seriously just because they show some correlation with real-world economic trends? Do they really have any predictive power?

In previous elections, streaks that at the time had correctly predicted, at least five times in a row, a victory by a specific party had a success rate of … exactly 50 percent! In other words, we spent our weekends building the world’s most elaborate coin-flipping machine.

To be sure, streaks that predicted whether the incumbent party would win — a method our friend Lenny Bronner says has some grounding in actual political science — had a slightly better track record at 53 percent. But that still falls in coin-flip range.

Perhaps longer streaks are more predictive, and the short streaks pull down the average? Not really! Streaks that “predicted” a Democrat or Republican winning had a roughly 50 percent success rate, whether they’d succeeded 11 straight times or six straight times.

Longer streaks regarding incumbents did fare slightly better: Streaks that had predicted 10 or 11 elections had a 56 percent shot at getting the next one right, while those with 12 or more correct predictions had a 62 percent shot. With certain data transformations, such as those that looked at whether the indicator had above-average monthly change, the success rate creeped still higher. But even those stronger-track-record streaks shrug at this election, giving Harris even odds, and most streaks proved useless.

Does that mean economic fundamentals can’t predict elections? No! It just means the methods we adopted from tabloids and social media — looking for an all-seeing data oracle — were doomed to fail, as we may or may not have expected.

To figure out the right way of going about this, we looked to <a href="https://split-ticket.org/">Split Ticket</a>, an invitation-only firm of five young professionals and students who enjoy spending their nights and weekends building a free data analysis and journalism outlet (they accept <a href="https://split-ticket.org/donate-to-split-ticket/">donations</a> but have no real revenue model). In other words, our type of people!

The founders met on Twitter, now known as X, and launched the site as a hobby. Three years later, it qualifies as critically acclaimed, perhaps even mainstream. Their work on the <a href="https://split-ticket.org/2024/08/22/a-very-detailed-examination-of-the-washington-primary/">national predictive powers of Washington state primaries</a> was <a href="https://www.washingtonpost.com/opinions/2024/08/19/kamala-harris-polls-democrats-washington-primary/">published by The Washington Post</a> and <a href="https://www.nytimes.com/2024/09/28/upshot/election-forecasting-washington-primary.html">covered by the New York Times</a>. They’ve also written for the <a href="https://www.nytimes.com/interactive/2024/05/24/opinion/ads-abortion-election-2024.html">Times </a>and <a href="https://www.politico.com/news/magazine/2024/08/21/kamala-harris-gains-polls-00175262">Politico</a>, and appeared on <a href="https://www.youtube.com/watch?v=uM4StnpXTmQ&amp;list=PL6XRrncXkMaWsHRkoIp3Q9WcQb4zhR6_t&amp;index=1">cable news</a>. Yet their grizzled leader, Lakshya Jain, is just 27 years old.

An AI engineer in the Bay Area by day, Jain did the data work underlying Split Ticket’s <a href="https://split-ticket.org/2024-presidential-polling-averages/">presidential</a>, <a href="https://split-ticket.org/senate-2024-ratings/">Senate</a> and <a href="https://split-ticket.org/house-2024-ratings/">House</a> forecasts, an effort that has earned props from the folks responsible for more established forecasts at Silver Bulletin and FiveThirtyEight. Of particular interest to us, they include a measure of economic fundamentals. (Jain and his colleague, Harrison Lavelle, <a href="https://split-ticket.org/2023/07/21/presidential-elections-and-the-economy/">laid it out in some detail </a>last year, if you’re curious.)

The original sin of the magic-number-style forecast, Jain explained, is that it boils down every election to a yes or no question. So if the Republican won with 51 percent, but a model said they’d lose with 49 percent, that counts as a failure while a model that said they’d get 71 percent would be counted as a success. But the “success” was off by 20 points, while the “failure” was only off by two.

A better model would predict the share of the vote and would focus on incumbency rather than Republicans or Democrats. After all, a strong economy isn’t inherently good for either party — it’s good for whatever party was in power and can (somewhat plausibly) claim credit for it.

And a good model won’t spit out a simple yes or no answer, like our tomatoes or cotton prices would. Instead, it’ll give a probability that communicates its confidence in the result: Split Ticket recently gave Harris a 53 percent chance of winning, an outcome they consider to be a toss-up.

Finally and most importantly, Jain explained the folly of this whole exercise: We have only about 16 presidential elections for which there’s a decent amount of economic data, which is not a long enough track record to separate outliers from the trend. You can torture a model until it has incredible accuracy on that tiny historical sample, but you’ll end up with a ludicrous, brittle set of assumptions that collapses in the face of anything unexpected — a fact we proved with all those streaks above.

So what does the team at Split Ticket do? They start with a handful of carefully chosen series — <a href="https://fred.stlouisfed.org/series/CORESTICKM159SFRBATL">inflation</a>,<a href="https://fred.stlouisfed.org/series/GDPC1"> economic growth</a> and <a href="https://fred.stlouisfed.org/series/UNRATE">unemployment</a>, for example — and test which combination comes closest to predicting the incumbent party’s vote share. That gives the model the best chance of succeeding in unpredictable circumstances, such as almost every presidential election.

“The purpose of this type of model … is never to get down to the exact decimal place,” Jain said. “The purpose is to get a rough understanding of where it was and get a ballpark approximation. That's all this indirect data can do. It can only give you an approximation of where things are supposed to lie.”

Still, their simple model, a regression of three variables, typically missed the incumbent vote share by just 1.2 percentage points in elections from 1972 to 2020. It struggled in 1972 and 1992 — math alone can’t account for George McGovern or H. Ross Perot — but was just 0.5 percentage points off in 2016.

This year, it would have given Harris an advantage of around a 2.2 or 2.3 percentage points in the popular vote — strikingly close to the advantage of 1.9 percentage points that she has in the Split Ticket polling average <a href="https://split-ticket.org/2024-presidential-polling-averages/">this week</a>. But Jain doesn’t recommend taking economic numbers too seriously.

Why? Because we have something better, a method of prediction whose powers Jain speaks of with something like reverence, a tool whose effectiveness that he describes as the single most surprising fact he’s learned in the election-prediction biz.

“People always like to say polls suck. And, I mean, that’s true to some degree,” Jain told us. “But polls are still the single best tool we have to gauge the state of play in public opinion.”

Bronner strongly agrees. He — and The Post — focus on <a href="https://www.washingtonpost.com/elections/interactive/2024/presidential-polling-averages/">polling averages</a> rather than odds-based forecasts, which he says are essentially impossible to prove. You’ll just never have enough presidential-election data to know if your 53 percent chance accurately describes the situation.

If you insist on doing a forecast, he and Jain agree that economic fundamentals make for a reasonable starting point. They should play a bigger role early in the election cycle, when voters know less about the candidates, accounting for more than a third of your forecast, Jain said. But as voters’ opinions about the economy merge with their views about specific candidates, Jain downgrades the influence fundamentals have on his results.

By Election Day, economics struggle to tell you anything not already contained in the polls — which turn out to be the closest thing we have to a real magic indicator.

Good day! The Department of Data constantly craves your quantifiable questions. What are you curious about: Is silk really making a <a href="https://books.google.com/ngrams/graph?content=silk&year_start=1800&year_end=2022&corpus=en&smoothing=3&case_insensitive=true">comeback</a> in a world of artificial fibers? Who’s most likely to have life insurance? Which jobs start their workday the earliest? <a href="https://thewashingtonpost.formstack.com/forms/department_of_data_questions">Just ask!</a>

If your question inspires a column, we’ll send you an official Department of Data button and ID card. This week, we’ll mail cards and buttons to Laura Dooley in Nampa, Idaho.

University of Oklahoma

University of Missouri

Andrew Van Dam writes the Department of Data column each week for The Washington Post. He has covered economics and wrangled data and graphics for The Post and the Wall Street Journal. 

Andrew Van Dam

the-washington-post

Getty Images/iStockphoto

Helios4Eos

Economy

Economic Policy

Personal Finance

Work

Technology

Business of Climate

Business

washpost

prism://prism.query/site-articles-only,/business&limit=20

Get a curated selection of 10 of our best stories in your inbox every weekend.

The latest business news, analysis and columns, including on the economy, personal finance, work, technology and more.

The problem with using the economy to predict the election