From ‘Wonder Woman’ to ‘Lady Bird’ to the Dora Milaje of ‘Black Panther’, complex female characters are appearing on the big screen. In music, female artists like Dua Lipa have been topping the UK charts. And the highest paid author? It’s J.K. Rowling. Yet many women still feel invisible in the arts - underpaid and under-represented. So how much has really changed? To properly answer this question we need to adopt more innovative ways of tracking gender balance.

Moving beyond surveys

Surveys are the most common method for measuring diversity in the arts. But surveys can be slow to deliver insights. From the time a survey is launched it can take several years before we see long-term trends. Moreover, surveys about gender tend to be limited to counting the number of men and women. They tell us little about differences in the prominence of those men and women, on the stage or screen. And they usually don't measure differences in how men and women were portrayed.

In recent years, new approaches have emerged. Rather than surveying respondents these studies infer gender using existing information. This information may simply be the end credits from a film, where an actor's first name reveals their gender. Or in a field like literature, the presence of ‘he’ or ‘she’ in the text of a novel can be enough to identify the gender of characters.

The arts are uniquely positioned to take advantage of these new approaches. Unlike other industries, their outputs are often public, and creators are named. From film credits to festival line-ups, these information sources can be mined to learn more about gender disparity in the arts.

There has been no meaningful improvement in the on-screen gender mix since the end of World War II.

Example 1: Using film credits to infer gender

Last year the British Film Institute (BFI) inferred the gender of over 250,000 cast and crew members from the credits of UK films released over the last 100 years. To do this, they not only used members’ first names, but also looked at clues offered by titles (e.g. Mrs and Mr) and character names (e.g. Uncle). Using their data, we were able to show that there has been no meaningful improvement in the on-screen gender mix since the end of World War II, and the percentage of women in on-screen roles still sits at around 30%. The peak for female representation in film (when women made up 41% of casts) was over 100 years ago, in 1917. These insights could never have been surfaced from surveys alone.

The BFI enriched their dataset by giving a unique identifier to each crew and cast member, which involved the laborious task of determining whether the “Mary Smith” credited in this film was the same “Mary Smith” credited in that other film. This allowed us to study actors' careers in the UK film industry. In successive decades, we found that men appeared in more films on average than women, and had a longer gap between their first and last films (a basic measure of career length). On a positive note, these gaps appear to be narrowing over time.

Example 2: Inferring gender by matching names across databases

Sometimes we don’t need to infer an artist’s gender; we can use existing databases instead. In 2016 we used this approach to study the gender mix of composers at the Proms, the world’s largest classical music festival which began in 1895. This analysis was only possible because the BBC keep an open archive of the 50,000 works that make up the Proms record. After matching composers’ names to a music database (which contained information on their gender) we showed that women often featured more frequently in the Proms during the early 20th century than they have in recent years. And according to a more recent study, women made up just 9 of the 120 composers who works appeared in the Prom’s 2017 season.

Measuring differences in prominence

By analysing artistic content directly, we can go beyond just counting the number of men and women and we can compare how prominently they feature. Researchers at Google, in collaboration with the Geena Davis Institute on Gender in Media, measured the time that actors spent on screen and the time they spent speaking. They focussed on the 100 highest grossing live-action films in the US for the years 2014, 2015 and 2016. Using facial recognition technology, they showed that men were seen and heard almost twice as often as women, with women occupying just 36% of screen time and 35% of speaking time.

More recently, researchers from the universities of Illinois and California at Berkeley compared the prominence of male and female characters in novels. They analysed the texts of more than 100,000 novels published over a 200-year period. In books written by men, women occupied on average just a quarter to one third of the character-space. In books by women, the division was much closer to being equal. The authors describe the gender gap as being “depressingly stable” over the 200-year period.  

Measuring differences in portrayal

Inferring gender can also be used to reveal differences in how men and are portrayed in the arts. In 2017 The Pudding compared words that follow ‘he’ and ‘she’ in the screen directions given to actors from around 2000 film scripts. They showed that women had a high likelihood (relative to men) to be asked to ‘snuggle’, ‘giggle’, ‘squeal’ and ‘sob’. Conversely men were more likely to ‘strap’, ‘gallop’, ‘shoot’, ‘howl’ and ‘kill’.

We also found evidence of gender stereotypes in our analysis of UK film. Over the last 100 years, the five unnamed roles most likely to be played by women, rather than men, were prostitute, housekeeper, nurse, secretary and receptionist. That top five doesn’t change even if we only consider films released from 1985 onwards. We also found that women were less likely than men to portray characters in highly skilled occupations, such as managers, solicitors and doctors. Since 2005 only 16% of unnamed doctors in UK films have been played by women, despite women now comprising 52% of doctors on the UK's General Practitioners Register.

The arts have the opportunity to lead the way in adopting more innovative measures of gender balance.

An opporturnity to innovate

This collection of new approaches shows how we can reuse information and content in novel ways to measure gender inequality in the arts. But these approaches are not yet commonplace and they tend to surface from one-off studies. That’s because gathering the necessary information takes time and expertise. It is no easy feat to assemble the credits from 10,000 films, or to collect the text from 100,000 novels. While it is perhaps not the most exciting task, careful curation of artistic content has been the key to realising several of the studies mentioned above.

Just like surveys, these new types of studies are not without their limitations. Databases are often incomplete and we cannot always accurately infer everyone’s gender from their name or face. For one, names may be gender-neutral, such as Mel or Alex. These studies also tend to presume that gender can only be male or female - a definite flaw. The solution is to openly discuss and measure these limitations.

With careful communication these studies can help to reduce our reliance on anecdotal evidence. All too often the latest blockbuster, novel or play is cited as ‘proof’ of change. To counter this we should aim to build an open evidence base, that relies not just on surveys but also utilises more novel approaches. The metrics should show long-term trends, and go beyond merely counting heads; they should capture disparities in the prominence and portrayal of women and men within the arts.

The arts have the opportunity to lead the way in adopting more innovative measures of gender balance. Now is the time to seize that opportunity.