In the April 11, 2024, episode of the 538 Politics Podcast, “What's up with the kids these days?” host Galen Druke, Split-Ticket.org partner Lakshya Jain, and Harvard University's Director of Polling John Della Volpe investigate recent trends in polls of the electoral politics of Generation Z. Specifically, they scrutinize polling data that appear to show that Donald Trump is gaining ground with young voters in the 2024 election, which contradicts previous narratives about the generation. As 538 is a media outlet that analyzes data, it's safe to assume those interested in the podcast will have some basic statistical knowledge, even if the podcast is directed to the general public.

As Druke points out in his introduction, these shifts have “prompted a number of observers to argue that the polls are all simply wrong” (01:14). I would argue that data and polls can't be “wrong” because they reflect some level of truth in what they record; it is the interpretation of that data that can be wrong. This is partly a semantic objection, but I raise it because it's worth it to wrestle with hidden assumptions of interpretations, and understanding that there is some kind of truth in these polls will help us better learn from them.

One question is about methodology: were polls conducted in such a way that leaves out certain demographics? Jain argues that Trump is not winning young voters, but rather, demographics who are more likely to support Trump are also more likely to complete polls and are thus overrepresented (04:34). Jain also claims that these figures are from general polls rather than polls that specifically target all kinds of young people, which makes them misleading. Druke also brings up that the 18–29 bracket may be too broad to identify patterns, as the 18–24 bracket is seemingly more conservative than the 25–34 bracket (11:20).

In absence of high-quality polls regarding voting, another main issue tackled in the podcast is regarding prediction: what other metrics can serve as indicators for voting patterns? Della Volpe argues that voting behavior boils down to value judgements (36:05), while Druke argues that knowledge and approval of policies, often targets of polls, don't matter unless the policies feel substantial or they personally affect voters (30:34). Della Volpe doesn't bring up any specific examples; Druke uses the example of Obama, a symbol of big change who won young voters by wide margins in 2008, and Reagan, a symbol of substantial conservative shifts who also won young voters in his elections. However, Druke fails to mention the almost endless confounding variables, or that these are different populations: the 18–29 bracket who voted in 1980 doesn't overlap with the 18–29 bracket who voted in 2008, and neither bracket overlaps with the 18–29 bracket who will vote in 2024.

Druke claims that the shift in support away from Biden and towards Trump is reflective of partisanship and self-identification, since fewer members of Gen Z than Millennials identify as liberal or Democrats (12:51). However, Della Volpe pushes back, saying that when looking at the issues, Gen Z is more progressive, despite not identifying as such (13:07). Their observations are correct, but they raise a question of whether partisanship or ideology is more predictive of their responses to the polls on their plans to vote, to which they don't cite specific figures.

The podcast almost exclusively focuses on expected values, almost never mentioning the standard deviations of these polls. The only time that variance is mentioned is by Druke, who suggested that the high variance among state polls represents high volatility among voters (44:24). This is not necessarily true, which can be understood with the Law of Total Variance. If the times that individuals change their voting plans are independent to their state or to each other, the volatility would cancel out with large sample sizes. Alternatively, failures to reach certain demographics, as mentioned earlier, may be more to blame for uncertainty. Because polls have to weigh the young vote more to make up for the lack of responses, this increases uncertainty in each individual poll.

Overall, I found the podcast interesting but vague. They fail to cite the complexities in the polling data, which is symptomatic of a larger problem in political media. They do raise interesting questions for researchers on the predictive power of various aspects of young voters on their voting behavior. They also problematize methodology to explain how data can paint misleading pictures in a changing polling landscape.