Towards a new era in the film industry?

Through the feminist struggles and the recent advances in gender equality, women have gained space and importance in movies.
Indeed, the cinema industry has long been a mirror reflecting societal norms and changes, and this is especially relevant for the representation of women.

Looking at the Cannes Film Festival, a prestigious sphere for cinematic excellence, change is palpable. This year, a striking 7 out of 21 films competing for the top prize are helmed by women (1/3). Cast your mind back to 1995, and the contrast is stark: a mere 2 out of 24 (1/12). This shift is more than just numbers; it's a testament to the slow but sure march towards gender parity in filmmaking.

Moreover, it is easy to see that movies, TV shows and other forms of art talk more and more about women realities, letting the archaic male-gaze in the past. However, like any great story, this one too has its conflicts. Stereotypes and inequalities stubbornly persist in the industry, reminding us that our quest for true equality is not over.

Let's embark on a journey of discovery together. In this study, we'll delve into the gendered dynamics at play in the film industry. Our guide? The CMU Movie Corpus dataset, a dataset of thousands of movies released between in the 20th and 21st centuries. To start, let us show you a first insight on the cleaned dataset we are working with, look at the graph below which tracks the evolution of the 9,146 movie releases through the years.

See? There are very few movies before 1980, this is why for the visual simplicity of future graphs which represent values per year, we will restrain the years range to the 30 years between 1980 and 2010 only.

Transitioning from our temporal exploration of movie releases to the geographical dimension, we uncover another layer of the film industry's story. Let's pivot our focus to the global map, where the geographical origins of these movies paint a vivid picture of industry trends and cultural dominance.

The map we showed highlights a clear bias: a vast majority of the movies between 1980-2010 originate from the United States and Europe. This concentration indicates a Western-centric dominance in the film industry, both in terms of quantity and cultural influence, or more likely a bias in our dataset. This geographical bias in our dataset, evident in the distribution of movie production, is a crucial factor to bear in mind as we proceed with our future analyses, ensuring a nuanced understanding of the film industry's landscape.

First, we could ask ourselves the following question: "Are there more women actresses who play characters in movies now than before?"

That is interesting! From 1980 to 2010, the proportion of women actresses compared to men increased of a few percents! It is not a huge increase, but it still is considerable. We'll have to investigate more on that and the potential causes for this increase, as well as whether when women are more represented, it also imply they are better represented or not.

Part I : What influences the presence of actress ?

In our journey to understand the role of women in the film industry, we first delve into the proportion of female actresses in movies. This exploration is only based on the data we have, without of correlating changes in the industry with broader historical events. A natural question arises: how do the roles of movie directors and producers influence this proportion? To answer this, let's look at the gender breakdown among these influential figures.

Observing the surprisingly low proportion of female directors, even more so than producers, prompts us to examine how these figures have evolved over the years.

Interestingly, the proportion of female directors has remained relatively constant, showing minimal change over time. But what about the producers?

Similarly, the trend for female producers mirrors that of directors — consistent but modest representation. This leads us to an intriguing aspect: is there a correlation between the presence of female directors and producers and the proportion of female actresses?

Against the backdrop of these low proportions, we find a silver lining. There appears to be a positive correlation between the proportion of actresses and the involvement of female directors or producers. Notably, when both the director and producer are women, the proportion of actresses tends to be even higher. Statistical t-tests confirm this correlation, revealing significant \(p\)-values of \(10^{-4}\) for female producers, \(1.76\times10{-23}\) for female directors, and \(2.66\times10{-15}\) when both roles are occupied by women. But there's more to the story. How does this representation vary across different movie genres?

Movie Genres and Gender Distribution

The plot above reveals significant variations in gender distribution across movie genres. Actresses are more prominent in genres like LGBT, Drama, and Romance, while they are less represented in Action or Crime movies.

The insights gained from our analysis paint a multifaceted picture of women's representation in movies. While female directors and producers are few, their presence correlates with a higher proportion of female actors. This correlation is further nuanced by the genre of the movie, indicating that the battle for gender balance in the film industry is influenced by various factors, including the type of movies being made. As we continue to advocate for more balanced and diverse representation in cinema, these findings underscore the importance of female leadership in shaping the narrative and presence of women on screen.

Part II : Representation matters but is it enough ?

Now is time to evaluate the quality of the representation of women in the film industry!

Actors age

First, we will look at the evolution of the age of male and female actors. Why? Simply because young women are generally prefered than old women, while we are supposed to appreciate men at every age.
Well well well... The median age of male actors remains higher than that of female actors throughout the observed period, reflecting a persistent age gap. This trend could be indicative of industry preferences or casting practices that tend to favor different age profiles for male and female roles. While actress are casted at the age of around 30 while the median for actors is 40. This does set very different standards contributing to the stereotype that age has more negatives consequences for the evaluation of women compared to men.

Actors roles

As we analyse the distribution of actress in movies we should also take a deep dive into what they play. As this may be very well contribute to stereotypes on screen.
We scraped the role played by thousands of actors in movies and classified their role into several categories. This helps us understand what kind of character they play.

Actors are credited in movies as several categories.
Some roles have names in the scenario, some only have job duties.


We translate theses categories to four level of importance, with more important characters having names and less charachter only being in a relationship with a named character or worse, with an unnamed character.
How does gender impact these different categories ?
Let's have a look at the roles credited as a job! We do see that overwelming majority of roles are played by actors for the two first categories. However, we observe that it's not the case for roles that are further away from the main characters have a higher number of actress. Let's look at what all of these roles are!
Our first question is: What's the most common jobs played by actors and actresses?
We find some roles which describe their position in the family. Theses roles were not classified by our NLP as supporting role to a named character as they may simply be label as "The grandmother" and not "Ted's grandmother". We observe that for the roles that are not describing their position in the family, actress have roles such as nurse, waitress, receptionist... while men seem to have higher status roles such as doctors, cops/sheriff, owner, reporter, priest... Let's take a closer look at who plays nurses and doctors. So doctors are overwhelmingly played by men while nurses are mostly played by women. How does it compare to reality ? Let's compare it to the 2008 U.S. Bureau of Labor Statistics. In the statistics, 68% of doctors are men, compared to 82% of unnamed doctors character's in movies. While for nurse, 90% of registered nurses are women, in movies it's 72%.

So, while not a perfectly identical to the gender gap in our society, we find a heavy bias in how the roles are casted. Let's check if this is evolving through time:
While we do see that movies in 2010s seem to completely differ from the other decades, this first look at the data seems to indicate less bias in recent years. Let's check how other professions are doing. Especially who plays jobs with a perceived higher social status ?

For instance: looking at the gender distribution of teachers and professors. One job title: professor is linked to higher education and seems to be a role heavily played by men, while the other: teacher is played by both actor and actresses. For a few of the jobs listed in the most common roles, we find that most of them are heavily gendered.

As we have a small number of roles played for each job, we will attempt to have a look at the bigger picture. Let's agregate some data using a useful metric.
If women play roles of jobs with a lower social status, what does their fictional average salary would look like? We used the data from the May 2022 National Occupational Employment and Wage Estimates to estimate the average salary of the jobs played by actors and actresses.

First thing we observe in the above distributions is an almost identical median for both genders! But we do observe some differences, as we can see that low-wages jobs are mostly acted by women while actors are casted to high paying jobs. We also find what we could call a glass ceiling as the maximum salary for jobs played by actress corresponds to the upper quartile of the men's job.
We will now take a peak at the 10 most played supporting role, credited in a relationship to a named character.
We see similar roles, unsurprisingly, most of them are family members to a named character but for the roles that are not, we notice again that they include lawyer and doctor for actor's role and secretary for the women's role...
So again, it's the same story here !

We will not look at the supporting roles credited as in a relationship with a person characterized by their job, as they represent too few examples.

So, what did we find ?
In summary, the contrast in the roles given to actors and actresses reflect common gender stereotypes in storytelling and media, where men are often associated with higher social status jobs, power, authority and action, while women are often portrayed as a relationship or play highly gendered jobs in their roles.
This is a useful piece of the puzzle on how stereotypes are conveyed in movies productions.

Bechdel Test

Imagine how different the cinematic landscape could be if more films met these criteria. What do you think? Do films that pass the Bechdel Test tend to resonate more with audiences, or is the impact more subtle, perhaps reflected in the gradual evolution of female roles in cinema? Let's delve into this intriguing topic together.

But first, an important consideration: we need to be cautious with those data, because not all movies from the CMU Dataset have been evaluated by the Bechdel test (only between 20 and 140 per year between 1980 and 2010). We should also consider that there is a bias on the set of movies that were entered by hand in this Bechdel dataset and also add that, while the Bechdel test can be a useful initial indicator of female presence in films, it is not a comprehensive measure of a film's portrayal of women.

So let's start this analysis, by looking at the evolution of the proportion of movies that passed the test:
There is quite an increase in the mean proportion of movies that pass the Bechdel test! (slope is 0.005)
Even though the proportion of movies that pass the test is still low, it is encouraging to see it increase over time.
Now, do you think that movies with higher Bechdel rating involve more women ?
Here, it is obvious that a higher Bechdel rating typically indicates a greater percentage of actresses (around 24% difference for movies which completely pass the Bechdel test compared to the ones that do not), this is a result we could reasonabily have expected, considering the test's inherent design. But still, let's not take any shortcut! It is not because women are present that they have interesting roles nor their own plot. So movies with higher number of actress tends to have a higher Bechdel rating.

Part III: From the Seats: Audience Perspectives.

But then, let’s answer the question: is it clickbait? Does it contribute to the popularity of the movie? Or even to the revenue?

Firstly, we'll take 2 metrics: the popularity of the movie, which is a score determined here by TMDB, and the rating of the movie.
There's an intriguing pattern in these glitzy displays: movies that only score 1 at the Bechdel test are often the most popular ones. Why is this so? The answers aren't straightforward. This could be multiple confounder into this complex tapestry. It could be that the film industry, with its influential might, chooses to shine a brighter spotlight on movies where male characters take the center stage, and where women only are... present. Or perhaps, the audience, like a river, naturally flows towards the stories echoing a significant male presence. So is it the chicken or the egg? Let's see if we can find any pattern between the percentage of actresses and the rating:

A friend to the rescue: Matching

Let's do something new ! Have you ever heard about Matching ? It's a relatively new concept in statistics, popularized by Donald Rubin in the 1980s. You can think of Matching as your trusty sidekick in reducing bias when estimating treatment effects in observational studies. Here's the challenge: for every treated unit in your study, you need to find one or more non-treated units. But not just any units will do. These counterparts must be similar in observable characteristics, creating a perfect balance against which the covariates are evenly matched.

Here, we will perform exact one-to-one matching, based on the main genre of the movie. By comparing movies within these genre-matched pairs, we hope to accurately assess the impact of genre on both Bechdel Ratings and audience popularity. This controlled comparison helps isolate the variable of genre, allowing for a clearer understanding of its role. We will have two groups

• Treatment Group: Movies from a genre that typically scores high on the Bechdel test.
• Control Group: Movies from a genre that typically scores low on the Bechdel test.

Doing this on our dataset, we get 880 pairs of matched movies! Let's see what we can learn from this. We want to know if the Bechdel rating plays a role in the popularity of a movie. So, let's try to estimate the following linear regression popularity ~ bechdel_rating. We obtain that

\[\texttt{popularity} = \beta_0 + \beta_1 \times \texttt{bechdel_rating} = 9.4 - 0.1140 \times \texttt{bechdel_rating}\]

It seems that the Bechdel rating and the popularity are negatively correlated. Interesting, isn't it ? In fact... not really. What is important to mention is also the \(p\)-value of these coefficients. It turns out that the \(p\)-value of the \(\texttt{bechdel_rating}\) is \(0.450\), which means that the value of \(\beta_1\) is not statistically significant. Therefore, we can not state that the Bechdel rating impacts a movie's popularity, and as a result, producers should not have any reason to be concerned with it.

Let's now look at the correlation directly between the Bechdel rating and the popularity, the box-office revenue and the vote average.

The heatmap on the right is showcasing the \(p\)-values linked with correlation coefficients from the left heatmap. All these values are between \(0.00\) and \(0.05\). This tiny detail is crucial. The patterns you see on the left heatmap aren't mere coincidences. They're statistically significant!

Let's focus back on the left heatmap. Here, the coefficients are quantified on a scale from -1 to 1. Imagine this scale as a spectrum. At one end, -1 signals a perfectly inverse correlation, like two dancers moving in opposite directions. At the other end, 1 suggests a perfectly direct correlation, akin to synchronized swimmers moving in harmony. And in the middle, 0 stands for no correlation, like independent artists painting in their own styles.

Now, here's the twist. Despite the beautiful array of data, the correlations are generally weak. This subtly hints that the Bechdel test ratings have a minimal direct impact on a movie’s financial success at the box office and its acceptance by the general public.

This means that producers and film directors allow themselves to not adhere to the Bechdel test as it does not influence them in any way, which would give no monetary benefits to make movies that pass the Bechdel test or even reducde difference in proportion between actresses and actors.

One step back: how about ethics ?

During all this analysis, we used an open source dataset without putting an emphasis on the ethical and moral responsibilities. Now, we need to navigate the complex ethical questions raised in our analysis with four guiding "ethical lenses": beneficence, non-maleficence, privacy, and fairness.

As we saw, our insights could potentially reshape how producers and directors approach actor recruitment. But this needs to be done carefully. Consider the Bechdel test – while its impact might seem limited, overlooking it could encourage sexist behaviors in the industry.

Now, turn to privacy. We're handling sensitive data, laden with personal details of actors. How do we protect their identities? The solution could lie in sophisticated techniques like \(k\)-anonymity or \(\epsilon\)-differential privacy, ensuring insights without invasion.

Finally, the lens of fairness invites you to ponder deeper. Our data analysis holds power – the power to influence perceptions and decisions. It's crucial to wield this power judiciously, ensuring our conclusions don't perpetuate discrimination, especially against women in the industry.

Conclusion

So what did we learn?

First and foremost, it’s not tomorrow that we will achieve gender equality in the acting world. We are far from a 50/50 ratio, and it may take decades.

We did observe that the proportion of actors/actresses in a movie was correlated with the gender of the director and the producer. This may be a good lead in order to fast track gender equality in the industry.

The deep analysis of the roles assigned to the characters allowed us to see that men and women generally play stereotypical roles. This does tell us that casting more actress is not enough! We need better roles in order not to reproduce patriarchal stereotypes.

We also saw that the Bechdel test give us something to look forward to, as movies seem to be passing increasingly!

Finally, we check how the audience felt about movies that passed or failed the Bechdel test and we did not find that more “feminist” movies are unpopular.

We hope that this analysis gave a first glimpse of the gendered dynamics at play in the film industry, and that it will encourage you to watch movies with a more critical eye.