Pundits love a political horse race, parsing the latest polls to predict who might win an election. And in the final runup to the US presidential contest, these forecasts can influence markets and shape public opinion and policies.
But as America heads to the polls in November, new research warns forecasters to use caution with one large body of data: past demographic voting trends. Analyzing census figures and US presidential and Congressional election results since 1952, researchers find that demographic factors considered reliable determinants of an election’s outcome—like voters’ age, race, gender, and education—may be less useful than forecasters believe.
“We find that even guessing at a 50-50 level is just as good, if not actually better, than using some demographic forecasts”
How bad are demographic trends in predicting election results? So bad that, in some cases, a forecaster would be better off flipping a coin to determine a probable winner. “We find that even guessing at a 50-50 level is just as good, if not actually better, than using some demographic forecasts,” says Vincent Pons, the Byron Wien Professor of Business Administration at Harvard Business School and one of the paper’s authors.
The main reason is that party and candidate positions on key issues subtly or not-so-subtly change from election to election in reaction to shifts in voter sentiment, undercutting the demographic trends. Political strategists hoping to make long-term plans based on voter demographics might need to rethink their approach.
“After an election, the electorate always seems to just oscillate back to the other party within a pretty short period of time,” explains coauthor Richard Calvo, a former research associate at HBS who’s now a doctoral student at the University of California, Berkeley. “Any prediction that says there’s going to be a long-term advantage to one party is inevitably going to be wrong.”
Pons and Calvo teamed with Jesse Shapiro, the George Gund Professor of Economics and Business Administration at HBS, on the National Bureau of Economic Research study, “Pitfalls of Demographic Forecasts of US Elections.”
The appeal of demographic forecasts
To be clear, the authors are not arguing that demographic factors are unimportant in elections.
In fact, demographic groups—whether defined by age, race, gender, income, education levels, or other factors—clearly have had very distinct voting patterns over the decades. Black Americans, for instance, have traditionally voted overwhelmingly for Democrats, while rural and working-class white Americans have increasingly thrown their support to Republicans in recent elections.
Demographics and corresponding voting data appeal to forecasters precisely because the available numbers are broad and accurate. Predicting long-term trends influenced by aging, migration, fertility, and mortality is possible. Even projections of the future size and composition of various demographic groups are often considered reliable, according to the paper.
However, assuming demographic voting trends are monolithic and impervious to slight—if not large—changes from one election to the next is where demographic forecasts routinely falter, the authors conclude. And the further out election forecasts go—say eight or 12 or more years—the more dubious the predictions become.
Studying almost 70 years of voting records and results
The team analyzed voting from the American National Election Study and demographic information from government sources going back to 1952. From there, the researchers:
- Created a predictive model relating a person’s vote to their age, gender, race, income, education, and the type of area in which they live.
- And used the demographic information to predict individual vote choices in the next election.
“We looked at how demographic groups may have changed in size and how they voted, say, eight years earlier,” says Calvo. “Then we (calculated) the expected election’s results based on those numbers.”
The results weren’t good for forecasters, even after the authors statistically “stacked the deck” to favor demographic-based election projections:
- Forecasts up to five elections in advance are about as good as guessing that the result will be the same as today’s outcome.
- Someone using demographics to predict election winners would be off, on average, by 22 percent more than someone who assumes the next election will always be a tie—a 50-50 split between Democrats and Republicans.
- Even a forecaster who knew future demographic trends perfectly would have poorly predicted upcoming elections.
“Using previous voting results and demographic trends and applying them to future elections is just not a reliable way to forecast upcoming elections,” says Shapiro.
Consider Trump on abortion and Harris on immigration
The findings reflect the constantly changing state of politics and voter behavior. In a separate piece of new research, Pons considers the array of voter participation barriers that have changed in recent years, including election laws, voting procedures, and registration requirements, and highlights questions for future study.
And when it comes to demographic-based election forecasts, both in models and reality: voter and party positions on issues are not static from election to election.
“Candidates and parties clearly pay attention to polls to understand which issues are important to voters”
Ultimately, political parties and candidates regularly shift positions to meet the changing views of various voting groups—and consequently, demographic voting patterns can change from one election to the next. Even the slightest percentage shift in votes by a demographic group from one election to the next can spoil a seemingly well-thought-out forecast, the authors conclude.
Such shifts are apparent in the race between former President Donald Trump and Vice President Kamala Harris.
Trump, a Republican, has recently been trying to soften his anti-abortion position amid signs he may be losing support from women. Meanwhile, Democrat candidate Harris has recently been toughening her stance on immigration and border controls, aligning with what polls say many Americans want.
“Candidates and parties clearly pay attention to polls to understand which issues are important to voters,” says Pons. “Trump is realizing that on abortion and Harris is realizing that on immigration. These candidates are trying to adjust to what the voters care about.”
You Might Also Like:
Feedback or ideas to share? Email the Working Knowledge team at hbswk@hbs.edu.
Image credit: Unsplash/MagicPattern
