Mini Post #7: I'm Just Ken
Mini Post #7: I'm Just Ken
Do sexist people like feminist movies?
I’ve decided to try something new: once a week, to write a shorter, narrower post focusing on one specific paper. Last week, I wrote about whether people buy fewer sandwiches if the guy selling them is a pedophile. This is the tab with all the previous (and current) posts.
Onto the actual post: do sexist people like feminists movies? And do an area’s traditions affect its taste in movies?
Celebrities and film buffs having a taste for obscure, European movies has been a topic of discussion in film Twitter and film TikTok for a while now. And while my own taste is fairly conventional (as per my Letterboxd), it does open the question: why do people expect others to like the movies they do? Why not just accept divergent tastes?. And, more generally, why do people like the films they like?
This week’s paper is the simply titled “Movies” by Stelios Michalopoulos and Christopher Rauh, published the day after the Oscars in fact. The paper tries to tackle two questions: the first, why certain movies are popular in different areas. The second question is whether this is because of the actual themes and content of the movie. Lastly, it looks into whether gender norms in each area play a role in whether the movie is successful or not.
The way they look at the topic is by assembling a database of ancestral stories, legends, and tales from all over the world, and gaming out the motifs and themes in each of them. Similarly, the authors develop a database of over 13,000 movies from IMDb from the 1990s to 2020, with each movie’s box office and details of its cast, crew, and distributors detailed for analysis. The idea the authors aim to test is simple: “When films incorporate elements and plots from a country’s folklore, they can resonate deeply with the audience from that culture. This resonance can create a sense of familiarity and connection, making the movie more appealing and relatable to the audience.”
One such example is the movie “Big Fish” by Tim Burton: the authors find that cultures that place heavy emphasis on father-son relationships (per their folklore) were more drawn to the movie than others. Other films and motifs, such as Final Destination (for motifs relating to death, and to fate) or Puss in Boots (for motifs about brains triumphing over brawls, especially involving animals), also track. To avoid selection issues, the authors leverage machine learning technology, to draw out matches between the anthropolgist’s list of motifs, and the plot summaries for each movie in IMDb. Using statistical techniques that are beyond the scope of the post (or my capacity for understanding them), the authors try to develop a dimensional set of indicators to match each movie’s plot to the set of motifs, obviously considering relative importance within the movie. To avoid this becoming tediously technical for both audience and writer, I’ll just say they develop indicators for both whether a movie’s themes are similar to the motifs of the area of interst, as well as one measuring the differences between a movie and a set of motifs. This, then, provides two alternative specifications. After doing this, the authors employ a simple strategy: first, develop an indicator that measures the distance between a given country’s average motif and a given movie’s average motif, across a series of potential topics that closely match qualities such as genre.
Now come the actual regressions and the points they try to make. First is the question of whether blockbuster films have more universal values; to make this point, they would have to find that the films with the highest box office revenue would also have very small distances to the average theme of a weighted average of all countries. This is done both across the entire sample, and for each year, using fixed effects for both countries and years. They also include controls for specific actors, directors, distributors, and genre, as well as consider that movie screenings are selected based on box office expectations. In both specifications, the desired relationship is found: there is a significant and negative relationship between the distance of a movie’s themes to globally recurrent themes, and its box office success. Similar results are found for regressions of whether a movie was screneed or not in a given country given its proximity to global motifs, and its distance from country-specific motifs. Thus, movies that match country motifs more closely are likelier to be successful in said country both in absolute terms and relative to other countries, and this extends globally: blockbusters have universal themes.
The authors utilize a similar strategy for the US in particular. First, they construct a list of 210 Designated Market Areas (DMAs), which are regions and subregions used by various industries and professional groups. They mapped Google Trends search interest for specific movies in both each year and the cumulative 2004 to 2020 period, and also compiled a dataset of motifs relevant to each DMA. The way they did this could be a bit weird: they put together a dataset of the ethnic composition of each DMA by nationality (so Jamaican, French, Russian, Japanese, etc) and then used those as weights for the motifs of each “origin” country. So a DMA that is 33% Latvian and 67% Liberian would have average motifs derived as a weighted average of Latvian motifs (with a weight of 33%) and of Liberian motifs (with a weight of 67%). For the US, they test the same hypothesis and the same regression: Google Trends popularity (ranked on an index from 100 to 0) as a function of the same controls, and of DMA weighted motifs. The results will not surprise: people in a market area showed more interest in movies that matched the motifs from stories close to the motifs of their ancestral homelands.
Next, the matter of interest is whether people watch movies that match their values closely. This is done by looking at two traits: risk aversion, and gender conservatism. Let’s start with risk: to measure entrepreneurial activities, they utilize patents per 100,000 people and the number of registered LLCs per 100,000 working-age people. For ancestral values, they utilize a set of tropes and motifs related to risky contests. The movies, meanwhile, are ranked by how ChatGPT would respond to the question: “How would an objective observer categorize the plot of movie ‘X’ directed by ‘Y’ in terms of risk-taking? Pick only one from the options below and provide an explanation (max 60 tokens): ‘Wins exceed Setbacks’, ‘Setbacks exceed Wins’, ‘Setbacks and Wins balance each other’, ‘No risk-taking behavior in the movie’, ‘Unknown movie’.” The specification is similar: box office success as a function of the interaction between the country’s relative risk aversion and whether ChatGPT ranked the movie as risk favorable or risk averse. Across both variables and using a series of controls and fixed effects, and for both box office and probability of screening, there is a significant and positive relationship between entrepreneurial values (both present and historic) and success in a country’s film market.
A similar strategy is employed for gender norms, where the proxy for national gender norms is female labor force participation, and utilize ratings by a different AI of how gender roles in each movie fit into different stereotypes, for both men and women. The authors estimate bias by measuring whether men or women in the movie are portrayed with positive traits associated with men (dominant, physically, active, and intelligent) or negative ones associated with women (naive, submissive, and engaged in domestic affairs). Subtracting the difference for each movie has the “bias” it shows: for example, the main characters of The Avengers (2012) are rated with the positive/masculine traits for both men (Iron Man, Captain America, Hulk, etc.) and women (Black Widow, that lady with a pixie cut), so the bias is zero. The regression, then, measures whether societies with low labor force participation enjoy movies that have negative bias (i.e. are sexist), versus societies with high labor force participation and non-sexist movies. The results, unsurprisingly, is that movies wirth positive female representation are more popular in “woke” nations, and that sexist movies are more popular in “conservative” nations - put another way, Swedes love Greta Gerwig’s Little Women, and Italians love Michael Bay’s Transformers.
In conclusion, and perhaps not surprising anyone, conservative cultures seem to dislike movies with un-conservative themes, and more open cultures enjoy those same movies - and vice versa. But we can all agree on one thing: Madame Web was terrible regardless of its gender politics.
To finish the post, some links
The paper in question
Two blog posts by me about the economics of Oscar-nominated movies
A blog post by me about the interplay between culture and economics
A blog post by Alice Evans about the link between ancient herding practices and female genital mutilation
A paper by Kearney and Levine about how the show Sixteen and Pregnant reduced teen pregnancy, and a contrary paper by Jaeger et al. that casts doubt on the same conclusion
A paper by Alesina et al about how the use of the plough in ancient times determines female labor force participation in the present, and a paper by the same authors linking the same phenomenon to lower fertility rates.