Roles and gender are changing in cinema and TV – but maybe not how you’d expect
Author: Matthias Gallé , Senior Scientist and Area Manager, Xerox Research Centre Europe
As anyone who likes watching classic movies can attest, films are a great way of getting a snapshot of an epoch. But they also mirror the real world, reflecting societal change, ongoing events and the debates of a period in time. They influence us and set expectations on behaviours, roles and attitudes. All of this makes movies (and TV) ideal objects for societal research which is what made them good objects of study for us at XRCE. We were particularly interested in how roles like those of a physician, secretary or president were depicted on-screen. To manually go through movies on a large scale is extremely time-consuming and not very realistic so we delved into our big-data toolbox to see if we could use some of them to get a broad but accurate picture.
We developed some large-scale text-mining algorithms to mine IMDb user-generated content of 18 million actor-role pairs (i.e. who played what) for over 50 years. We focussed on analysing the evolution of roles over time and the gender of who portrayed them. This basically allowed us to see for instance which roles where prominently played by males and which by females. To cite some unsurprising examples, maid and receptionist are frequent roles which are mostly female, as are belly dancer, stripper and cheerleader. On the male side, there seems to be strong bias for referee, doctor and lawyer together with some criminal or negative roles (rapist, terrorist, thief, thug and a series of security or military roles (US. soldier, police officer ‘cop’, general). Also worth noting is that, whilst ‘therapist’ is gender neutral, psychiatrists are moderately male (are played by male actors between 60% and 80% of the time) and psychotherapists moderately female. Whilst a psychic is moderately female, a paranormal investigator is moderately male. In gender neutral, we find swimmer, student, church member and obstetrician.
More examples of this kind of stuff are shown in the table below, where p(F) represents the proportion of female actresses portraying that role:
We can combine similar roles to get overall statistics for a profession:
Combining this with the temporal dimension lets us see how the gender of these roles has evolved over time, with plots showing the changes, for example, in the gender representation of ‘nurse’. In general there is an upward trend of depicting nurses on-screen, but this trend is faster for actors than actresses. In 1990 around 95% of all nurses were female, while in 2015 this was just over 80%:
We also matched those roles to the reality as reported by US censuses, and compared how well the gender distribution on-screen matches reality:
Intuitively, points on the diagonal line have an onscreen portrayal consistent with the census distribution (“OES, for Occupational Employment Statistics). If a point is above the line (e.g. reporter), then those roles are overrepresented onscreen by female performers. Conversely, points below the line suggest an underrepresentation onscreen by female performers. For example, surgeons, teachers and nurses are played more frequently by male performers than their real counterparts.
Our research shows that this user-generated content can be used to gain interesting insights into popular representations of roles… assuming you have the right big-data mining tools. This can complement important societal research into existing problems like the gender gap and how popular views of those roles changed over time . It could ultimately not only cast light on those problems, but also inform writers, studios and film-makers about the impact of their choice, and about extreme cases of misrepresented roles and the associated implications.
You can find many more findings, and all the details, together with pointers to replicate the experiments by yourself in our paper Tracking onscreen gender and role bias over time, W. Radford, M. Gallé - The Journal of Web Science , Vol 2, No 1 (2016)