First the good news: Over time, as an election gets closer, polls often come into greater alignment in measuring the preferences of important subgroups in the electorate. The bad news: Most national polls may have large enough samples to offer real information on the standing of candidates overall, but they rarely include enough respondents to provide reliable information on key constituencies. This does not stop analysts from picking up on what seem to be dramatic subgroup findings and using them to offer sweeping conclusions, even if the results are based on a remarkably small number of interviews.
My Brookings colleague Gabriel Sanchez made this point about polling of Latino voters in a pair of articles published in May. His conclusion is right and important:
Most national polls are designed to provide a snapshot of the full electorate’s attitudes, voting preferences, policy priorities, and enthusiasm levels at this stage in the campaign season. Reputable polls do a good job of meeting this goal. However, these national samples are not designed to capture variation within sub-samples of the larger electorate, including Latino voters. Consequently, many national polls either suffer from a low sample size of Latino voters and/or a sub-sample of Latinos that is not designed to be nationally representative of that population.
The same is true for another hugely important 2024 voter group: the young. Voters under 30 (or, as some polls measure the young, under 35) are closely watched because to win these days, Democratic candidates need big margins among the young—and also need them to turn out in substantial numbers. Partly because our electoral system better fits the needs of older than of younger voters—the young tend to be more mobile and thus need to re-register more often—young people usually (though not always) cast ballots at lower rates than the older voters.
Start with the problem of measuring the preferences of young people. In early September, the New York Times/Siena poll and the Pew Research Center’s poll were released within a day of each other and covered roughly the same polling period. The Times/Sienna survey was conducted from September 3 to September 6; Pew from August 26 to September 2. They produced very similar topline results. Among registered voters, Times/Sienna showed Trump up by two points among registered voters while Pew had the race tied. But their findings for young voters were widely divergent, as the chart here shows. In the Times/Sienna poll, Harris led Trump by just one point, 47% to 46%, among voters under 30. In the Pew Poll, Harris overwhelmed Trump among the under 30s by 29 points—63% to 34%. This pattern of wide divergence was not confined to these two surveys, but it was particularly striking given the similarity of their overall numbers.
Over the course of September, the two surveys came more closely into line with each other, with Pew’s earlier measurement of Harris’ advantage among young people looking closer to an accurate estimate. In late September/early October, polling over very nearly the same interview period, Times/Sienna gave Harris a 58% to 37% lead among the young; Pew registered a 57% to 34% Harris youth advantage. This is broadly in line with other surveys during the month, including the gold standard Harvard Institute of Politics Poll, based on a large sample focused only on young people. Similarly, a Yale University youth poll released in mid-October showed Harris ahead among under 30s, 56% to 35%.
To get accurate readings of the preferences of important subgroups, Democratic pollster Anna Greenberg recommends surveys directed to that purpose.
“Generally, stand-alone polls of hard-to-reach groups are better because they are designed specifically to reach them and better reflect the diversity of these groups,” Greenberg said, referring specifically to polls of young voters and Latino voters. Young people, she said, “are less likely to answer the phone and take surveys than older voters and new methods like SMS [Short Message Service] polls do not do better with younger people.”
Younger voters are also more racially diverse than other generations and increasingly polarized along gender lines. In a relatively small subgroup sample, differences between the precise type of young people that a given poll reaches can thus add up to produce strongly divergent results.
Whit Ayres, a Republican pollster, said that to make their polls representative, pollsters sometimes use quota methods to cover race, gender, and education levels. However, other subgroups are not included in such quotas. “I suspect if you weighted these age cohorts by gender and education, they would be closer” to each other, Ayres added.
The bottom line: As a national poll is broken down into smaller and smaller subgroups, numbers from those groups are increasingly less reliable.
If measuring young people’s preferences is hard, figuring out their turnout rates—and, as importantly, which younger people will actually cast a ballot—is even more difficult. And exactly who among the young votes could matter enormously. This finding from a Pew survey conducted in late September and early October brought home the importance of differential turnout between supporters of the two major candidates.
“As in 2020, voters under age 30 are the least likely to say they are motivated to vote: 40% of voters in this age group say they are extremely motivated to vote,” Pew found. “And Trump supporters under 30 are less likely than young Harris backers to be highly motivated to vote (34% to 50%, respectively).” If this motivational difference holds up between now and election day, Harris could outperform even accurate measures of the preferences of the under 30s.
Young voters were critical to Democratic victories in 2018, 2020, and 2022. They could prove to be just as important in 2024. So, analysts will continue to watch their behavior closely. The best advice, offered recently by Ezra Klein in the New York Times and Eugene Robinson in the Washington Post, is for everyone to stop obsessing over day-to-day changes in the polling. But since most political junkies will ignore this counsel, they should at least avoid cherry-picking subgroup results that seem most startling or confirm their prefabricated assumptions. Comparing many surveys of subgroups before reaching large conclusions—and keeping a close eye on who is most likely to cast a ballot—provides a path to wisdom, and lower blood pressure.
Commentary
If you must be a poll junkie, be wary: The case of the youth vote
October 17, 2024