This is a perfect encapsulation of how we’re all feeling: unsettled. The presidential race has been neck-and-neck for months, even before President Biden decided to step off the ticket back in July.
As the 2024 general election approaches, we thought we’d dig into some maps and visualizations of election data from across the media landscape that help us understand the state of our union. In an effort to keep it positive, we’re going to focus on visuals that beautifully capture and explain complex ideas, infused with some much-needed humor along the way.
There’s a lot to look at here, so we’ll handle this topic in small bites. Today, we’re talking all about polling and its drawbacks. We’ll continue this series with part two (election indicators) tomorrow and part three (electoral math) on Friday.
What goes into polling?
Pollsters seek to get a sense of current public opinion from a representative sample of the voting population (or a subset of that population). Polling can be misleading. People are very quick to trust numbers without understanding what they actually mean or what demographic groups they may represent. At this stage in the campaign, most of the “information” we have about the state of the race comes from polling numbers.
Polling generally uses outdated methods of data collection, like calling people on the phone, that can skew the results in favor of certain demographics over others (just think: when is the last time you answered a phone call from a random number?). Many people prefer not to talk about politics unless they have to, especially in our current extremely polarized political climate. Put another way:
It’s difficult to get an accurate reading of how people feel about a complex set of issues when most people won’t respond to the question at all (MSNBC has some harrowing stats about this).
Knowing how difficult it is to engage voters in polling in the first place, Reuters recently dug into how polling results can lack the full picture. This visual is a very effective way of explaining how polling can provide information, but the level of detail varies considerably. As you scroll down the page, the number of pixels in the visualization increases, as does the detail in the image.
This is a lovely use of “scrollytelling” as we like to call it – navigating through a visualization by simply scrolling down the page. As a reader, I’m immersed in the experience from the beginning.
Reuters goes on to explain how reflecting an accurate breakdown of the population in sample data can make a huge difference in the level of detail a poll returns.
In the example below, a poll that includes an unrepresentative sample of education backgrounds returns some information, but the full picture is pretty unclear. If you adjust the sliders to match the actual breakdown of education demographics in the U.S., the picture becomes clearer.
These visuals really cut to the core of the problem in a way that is simple to understand. As is true in most things in life, the outputs of a poll are only as good as its inputs.
So how can we know if a poll is accurate? There are groups (here’s one example) that rate pollsters based on their methodologies, sample sizes, and historic accuracy. Reuters suggests that looking at lots of polls is another good way to get a more complete picture. Thankfully, FiveThirtyEight suggests that polls in this election are likely of a higher quality than in recent election cycles.
Polls certainly aren’t perfect, but it’s great to see improvements!
Visualizing uncertainty
With that in mind, how do we make sense of the data we do have, in a way that contextualizes doubt and uncertainty? These lollipop-ish charts from The New York Times nicely capture how often polls “miss” the actual outcome of a presidential election, highlighting the party whose candidate they underestimated. As you can see, missed polls can favor candidates from either party. Note that these are showing the popular vote margin, which is absolutely not the whole story (more on this in part three).
According to The Times, state polls tend to be even more off than national polls, which is significant in elections (like this one) where only a handful of states have competitive races.
I can’t help but notice that most of the swing states for our current cycle – Michigan, Nevada, Pennsylvania, Wisconsin, Arizona, Georgia, and North Carolina – fall into the underestimated for the Republican candidate category while the estimated result had the Democratic candidate winning in 2020 and many also in 2016. Seeing that there were five (!) swing states projected for Democrats won by Republicans in 2016 helps explain why the results felt like such a shock to many people.
In real time
This data was all about previous elections, which is often how we place the current cycle in context. For a look at the 2024 race, The Washington Post has a real-time polling tracker that helps us understand 1) the state of the race and 2) how things have changed in the last week.
This visualization is simple, but that’s why it’s successful. It’s very clear who is leading in each swing state, how much they’re leading by, and whether the other candidate is gaining any ground. This chart boils down the complexity of the election to the seven states that mathematically matter, especially during this race-to-the-finish in the last two weeks of the election cycle.
The New York Times has another nice breakdown of state polling, which follows the same seven swing states over the last several months. We love small multiples and they are very effective here!
Comparing swing states to each other quickly illustrates how different the races are in each of these states. As we know, an election is almost never as simple as one or two key issues. This also really underscores how little we should care about national polls! A one point lead in national polls doesn’t mean much in the world of the electoral college. Seeing the range of polling results over the course of the election is a nice touch here, too.
While looking at these state-by-state charts, I found myself trying to remember (and subsequently, googling) “what happened around August 28?” to understand why there was such a big swing across a few states – thankfully, NYT already thought of that. If you scroll down a bit on that page, it includes a nice, simple visual of how the lead has changed relative to several key moments from this election cycle.
The big swing was right after the Democratic convention and Harris has kept the lead in Nevada since then (with another boost after the debate).
Silver Bulletin, Nate Silver’s new blog after leaving FiveThirtyEight last year, includes a similar visualization for polling at both the national and swing state level since the summer. This chart does it all in one, with third-party candidate RFK Jr. included as well (and, as we saw in 2000, third parties matter in close presidential elections!). Note that Silver uses a different set of polls than what WaPo uses, hence the slight difference in trends and values.
A big part of effective design is anticipating the questions your audience will ask when they engage with your visualization. We love when the answers are already there.
Tomorrow, we’ll continue this series with a look at indicators of election outcomes, both past and present. Friday, we’ll wrap up with a look at visualizations that help explain the Electoral College.
Hope you’ll join us then! And if you haven’t already, make sure you’re registered to vote and have a plan for Election Day.