“Tiger King” fandom is correlated with voting Republican—and other connections between TV & politics
Last year, I came across this 2016 Upshot article from the New York Times. Besides being an example of beautiful data storytelling, the premise of the content intrigued me. Is there actually a connection between what TV shows people like, and how they vote in elections? NYT claims so. But as much as I loved the article, I was left wanting more. The article never went into detail on the actual correlations between TV and politics. So I decided to fill in some of that gap — and do so from the lens of our most recent presidential election, of 2020.
The New York Times article used data from Facebook page “likes” for TV shows. I do not have access (or haven’t figured out how to get access) to this, so I went with a different data source as my proxy for TV show preferences: Google Trends. Google Trends allows users to see how “popular” a particular search term or topic is in a given geography over a given period of time. So if I type in Queen’s Gambit, the relatively new hit Netflix miniseries (by the way what makes something a “miniseries” versus a normal series?), I can see all types of information about the show. Like, for example, how its popularity has evolved over time, as measured by volume of Google searches:
The show came out on October 23rd, 2020 — right where that peak happened. Makes sense. We can also see where the show was most “popular” (again, measured via Google searches) over the past year:
It looks like within the U.S., Washington, DC and Vermont are particularly fond of the show, or at least fond of Googling the show. North Dakota, on the other hand, doesn’t care.
And here’s the popularity of Queen’s Gambit alongside the popularity of just “chess” as a search term over time. In this case, over the past 5 years. The blue line is Queen’s Gambit; the red line is “chess.”
Correlation doesn’t mean causation in data, but here, I kinda sorta would say that it does. CNN reveals that “in the first three weeks after [the show’s] debut, sales of chess sets went up by 87% in the US and sales of books about chess leaped 603%.” Not surprisingly, TV shows and other cultural productions can have very real, tangible connections to people’s other behaviors. Not necessarily causal, but powerful nonetheless. Hence the topic of this blog post.
To tackle how TV show preferences and election choices relate to one another, I started by pulling Google Trends data across U.S. locations for each of the top TV shows from 2020.
My “top TV shows” list is a combination of the top 5 most-searched-for shows in 2020 according to Google Trends, and the top 5 regularly-scheduled TV programs of 2020 (with sports removed) according to the research firm Nielsen. And then I threw in Battlestar Galactica and Schitt’s Creek too for schitts and giggles (sorry). (Well, actually because I am personally a big fan of BSG and because I was surprised that Schitt’s Creek, which won 9 Emmy’s awards this year, didn’t make it onto the Top 5 lists mentioned above.) All this resulted in the following list of 12 shows:
- Tiger King
- Cobra Kai
- The Umbrella Academy
- The Queen’s Gambit
- Blue Bloods
- Chicago PD
- Young Sheldon
- Battlestar Galactica
- Schitt’s Creek
** Note: Chicago Fire actually ranked above Chicago PD on Nielsen’s list. I skipped it in my analysis because of oddly sparse data about the show in Google Trends.
From there, I pulled in data on how each county voted in the 2020 U.S. presidential election, scraped from the New York Times, Politico, and Fox News by this helpful person.
And then, armed with nothing but Tableau (a lot of it), some Google Sheets (for real), and coffee (decaf), I got to work.
Unfortunately Medium doesn’t currently allow for embeds of interactive Tableau charts, so all the visualizations in this blog post are screenshots. However, you can find the original, interactive version of all the charts and maps below here, in case you want to play with them more.
But first, how different parts of the country voted
If you paid any attention to the presidential election we just had, you’ve probably seen some version of this map below. I made this one using the county-level election data I mentioned. Darker red means a higher percentage of that county voted for Trump. Lighter red means the county still went Republican overall (more people voted for Trump than otherwise), but by smaller margins. Same concept goes for the blues, for counties that went for Biden.
So much red! Any map like this is, of course, misleading. Lands don’t vote. People vote. Joe Biden won ~51.4% of the popular vote, versus Donald Trump’s ~46.9% — a difference of more than 7 million votes. That’s why I like what the web comic xkcd created: an election map oriented around population instead of land.
Now, to the question that this project set out to answer: are TV show preferences correlated with voting behavior? It turns out the answer is actually yes, much more so than I ever expected.
Shows correlated with voting Republican
Counties where the following shows are more popular, are also more likely to have gone Republican in the 2020 presidential election.
These are in descending order of “slope,” or the difference in show popularity between the most Trump-voting counties and the most Biden-voting counties (on a scale from 0 to 100 due to how Google Trends scores popularity). Young Sheldon wins as the TV show on the list where Republicans and Democrats differ the most — the most heavily Republican counties apparently like this show almost 24% more than the most heavily Democratic counties. And Tiger King is almost 14% more popular among Republican-heavy counties than Democrat-heavy counties.
As for “p-value”: a p-value of less than 0.05 usually means the correlation between two variables is statistically significant (i.e. unlikely due to just chance). Note that the p-value for every show in the table above is much less than 0.05. In other words, there is highly significant correlation between how much a county likes these shows and which candidate won that county in the 2020 presidential election.
And below is a show-by-show view. The maps illustrate where in the U.S. a show is most popular (darker shade) vs least popular (lighter shade). The scatterplots — and the trend lines through them—show the relationship between percentage of votes that went Republican (x-axis) and show popularity according to Google Trends (y-axis). Each dot on the scatterplot represents a county, and the size of the dot reflects the relative population size of the county, approximated using the number of total votes cast in the 2020 presidential election (however, the correlation analysis itself is population agnostic — meaning each county counts equally regardless of population).
(Click on images on enlarge)
- You may have noticed that I labeled a small number of counties in the scatterplots above, in order to call out a few reference points in the sea of dots. Many of these are counties I have personal connections to, for example Alameda County (where Oakland and Berkeley are located) and Cook County (where Chicago is located). Others are there for contrast. These include: Macon County (the most liberal county in Alabama, with only 17.67% of votes going to Trump, less than even Los Angeles County, Cook County, or Alameda County), Lassen County (the most conservative county in California, with 74.84% of votes going to Trump), and the 3 most Republican counties in the whole country (Roberts County, Borden County, and King County — all in Texas and all with 94%+ of votes going to Trump).
- From the scatterplots, it’s clear how much bigger — and fewer — the blue dots are compared to the red dots. It’s further visual representation that there are more Republican-leaning counties overall in the U.S., but the Democrat-leaning counties tend to be much more populous.
- Many Americans, particularly in conservative-leaning places, seem to love cop/detective/crime shows. Of Nielsen’s top 5 regularly-scheduled TV programs, 4 are in this genre, and 3 of those 4 lean Republican (the 4th one, Chicago PD, doesn’t lean either direction, as I’ll get to below).
- It goes without saying but the correlations across this blog post are generalizations. As shown in the scatterplots, there are definitely Democrat-leaning counties that love these Republican-correlated shows, and Republican-leaning counties that hate these Republican-correlated shows. There are also most likely plenty of Tiger King lovers within any Tiger King -hating county, plenty of Tiger King lovers who voted for Joe Biden, and so forth. They’re just not as common, but they exist. This analysis is on the aggregate level and there are certainly individual variations within the aggregate.
Shows correlated with voting Democrat
On the other side of the aisle, we have these 5 shows that are all correlated in a statistically significant way with voting for Joe Biden.
Schitt’s Creek is the TV show that sets Democrats apart the most from Republicans. The Rose family is almost 26% more popular in Biden-dominant counties than Trump-dominant counties. And while Queen’s Gambit has single-handedly boosted the popularity of chess, the show’s fame is not equally distributed: the most Democratic counties like the miniseries almost 18.5% more, compared to their Republican counterparts.
Umbrella Academy and Ozark both lean Dem, too, but not by much. The correlation is statistically significant (again, p-value < 0.05), but there is a mere 2–3% difference in how much Trump voters and Biden voters seem to like these two shows.
Here’s the show-by-show view:
Shows with no significant correlations
Cobra Kai and Chicago PD are the only two shows on my list that have no statistically significant correlations to presidential voting behaviors.
As seen in the scatterplot, Cobra Kai leans sliiiiiightly Democrat (by 0.89%), but the p-value is 0.3448, AKA not statistically significant (too high that the correlation could be simply due to random chance).
Also, looking at the map: clearly counties in Southern California and much of the American Southwest love this show. The show’s setting in Los Angeles may have something to do with it.
Chicago PD is really really popular in the state of Illinois. This is utterly unsurprising to me: when I lived in Chicago, I was consistently impressed by how much pride Chicagoans had in their city and therefore how much they loved anything related to Chicago. To control for this, I made two scatterplots. The first one (middle image above) retains all counties, like all the other scatterplots in this blog post. The second one (right image) removes counties in Illinois. This made a bit of a difference, but not enough to change the story. In both cases, the correlation is not statistically significant (0.9119 with Illinois, 0.4204 without Illinois), and the show popularity difference between Republican and Democratic counties is negligible (0.17% with Illinois and 1.22% without Illinois, leaning Republican).
Bringin’ it all together
I expected TV preferences and political preferences to be related. I did not quite expect the correlations to be this significant or widespread.
Here’s how our top 12 shows stack together, in descending order of how much it divides Republicans and Democrats:
So apparently, your choices in TV shows can say a lot about your politics. Next time you want to fly your partisan flag loud and proud, suggest Schitt’s Creek or Young Sheldon as the evening entertainment. If, however, you want to avoid stirring the pot at your Thanksgiving family gathering, put on Chicago PD or Cobra Kai.
And of course, if your goal is to maximize looks of bewilderment, you should stream Tiger King. Obviously.
Meta reflections from doing this data project
- I used Tableau more heavily in this project than any other, and was reminded of how powerful the software really is.
- I’ve done correlation analyses in previous data projects, but not to the extent of this one. I had to brush up on some foundational statistical and mathematical concepts (like slope and p-value) in the process. Big thanks to my friend Luis for all his expert help here.
- Your data creates limits to your analysis. My decision to use Google Trends in this project meant that I had to live with the consequences, so to speak. For example, Google Trends data (at least the public version) only drills down to the metro area level, not the county level. This means that my show popularity data is less precise than my election data, because each metro area usually encompasses multiple counties. This is why the maps in this blog post look a bit “blocky.”
- I did this project on the aggregate level (metro area and county) because my data was on the aggregate level. But this does mean that some individuals’ behaviors are weighted much more than others’ in the results. For example, the TV preferences of rural or smaller areas likely got drowned out in the aggregate results of their larger metro area. And the smaller (less populous) a county is, the more sway each person proportionally has over that county’s election results.
- Because you can use Google Trends to get popularity data for literally anything, the analytical framework of this project is actually applicable to any topic. You can use it to figure out the correlation between anything and p̶r̶e̶s̶i̶d̶e̶n̶t̶i̶a̶l̶ ̶v̶o̶t̶i̶n̶g̶ ̶b̶e̶h̶a̶v̶i̶o̶r̶s̶ anything else with geographic data. For example, hot dogs. Apparently, hot dogs are the food of the liberals. Biden-loving counties like hot dogs 4.3% more than Trump-loving counties. And yes, that correlation is statistically significant (p = 0.0001).
I wanted to come up with a clever wrap-up to this blog post, but honestly, there’s nothing that tops hot dogs. (Except for mustard and ketchup and sometimes relish if you’re in the mood.) So I wish you a good day. May you entertain yourself this evening with a TV show true to your political self.