Stereotypes on screen

What do 3 million plot summaries, IMDb's full catalogue and TMDb's records say about how film and TV cast, write about and label women and social groups? Thirteen studies, 1920–2026.

Content note: this is about stereotypes, so it discusses them. Slurs and insulting labels are counted in the data but never shown.

The 0.95 that wasn't

In 2022, Stefano Li Pira and I trained a BERT model to spot stereotypes in plot summaries, and it scored an impressive AUC of 0.95. Its positives were long film synopses and its negatives short episode plots, though, so text length alone could tell them apart.

In 2026 I went back and rebuilt it properly, with matched comparisons and series kept whole in cross-validation. The same idea then reaches 0.71–0.76, and what it detects is stories about prejudice, not stereotyped portrayals. So I started again: 3,000 summaries labelled against a written codebook, modern open embedding models, and measures checked against those labels. Here is what came out.

01 · who plays whom

Women got the detective jobs. Men didn't get the nurse jobs.

Detectives, judges and lawyers went from almost never played by women before 1970 to 20–30% women today. Nurses, nannies and receptionists are still 83–90% women. Change came from women moving into roles coded male, not the other way round.

Casting skew by job, 2010–26 against 1920–69: jobs such as detective, judge and lawyer moved towards women, while nurse, nanny and receptionist stayed strongly female. — Casting skew by job: women's share of a job against their share of all named roles.

02 · who writes it

Films written by women cast more women.

Films written only by women cast 11 points more women than comparable films written only by men, and 8 points more even among the same director's films. Within a TV series the gap is about 3 points, which fits writers working inside a series' fixed cast.

Differences in the share of women in the cast and in the first-listed role for films and episodes written by women, with 95% intervals. — Casting differs with who writes: films and TV episodes, 95% intervals.

03 · how summaries talk about them

Half the characters now, still described by their looks.

Women went from 28% to 46% of the people plot summaries talk about. They are still described by their looks 1.5 times as often as their share would predict, down from 3 times in the 1950s.

How often women are described or act in each category, relative to their share of mentions, by decade; appearance stays highest. — Women's share of each kind of mention, divided by their share of all mentions, by decade.

04 · the trend

Stereotyped portrayals are fading, mostly the gendered ones.

Trained on those labels, an EmbeddingGemma-2 classifier recognises stereotyped portrayals with AUC 0.86–0.90. On 37,049 fresh summaries, film portrayals fall from 22% in the 1930s to 8% now, and on TV from 15% in the 1950s to 5%. Almost all of the fall is in gender stereotypes.

Share of film and TV summaries portraying a stereotype by decade, falling from about 22% to 8% for films. — Stereotypes by decade in a fresh random sample of summaries.

05 · behind the camera

Directors: from 3% women to 16%.

Women went from 3% of film directors before 1950 to 16% in the 2020s. They still hold only 38% of first-listed film roles, and in the same film the first-listed man is about six years older than the first-listed woman.

Women's share of directing, writing, producing, editing, cinematography, music, production design and casting credits by decade, for film and TV. — Women's share of jobs behind the camera, fiction film and TV.

06 · money, critics and awards

Stereotyped films earn less.

A film whose summary the classifier reads as a stereotyped portrayal takes 23% less at the box office than the same director's other films at the same budget, and is 2.2 points less likely to get an Oscar nomination (about 1 rated film in 85 gets one). Critics and audiences mark these films down alike.

Box office and Oscar-nomination differences for stereotyped portrayals, women in the cast and the six groups, with 95% intervals. — Box office at the same budget, and Oscar nominations, 95% intervals.

07 · who makes them

Women behind the camera, fewer group stereotypes.

Films and series made by women contain 15–34% fewer group stereotypes in the main comparisons, and no more gender stereotypes once their casts are compared.

Differences in stereotyped portrayals for films directed or written by women and series created by women, with 95% intervals. — Women behind the camera and stereotyped portrayals, 95% intervals.

08 · six more groups

Some stereotypes never left.

Immigrants, disabled people, people with a mental illness, and older, poor and homeless people are stereotyped in a steady 5–6% of film summaries from the 1930s to the 2010s (4.5% in the 2020s). Unlike gender, there is no long decline.

How, and how far to trust it

IMDb's datasets, TMDb's API and Wikipedia's episode pages; parsing with spaCy, word embeddings, and classifiers on open embedding models trained on labels written to a codebook. Titles are compared within a series, a director's films, a franchise or similar films, with summary length held fixed. Only aggregates are published: no plot text, no per-title data.

These are associations, not effects. Summaries are written by contributors, not the productions. The labels come from one language model following a written codebook; a second pass by the same model agrees closely, which shows consistency, not that independent readers would agree. All thirteen studies, the code and the caveats are in the repository.

This expands Stefano Li Pira's 2022 initiative to measure stereotypes in TV series, which we first explored together.

Information courtesy of IMDb (https://www.imdb.com). Used with permission.

This product uses the TMDB API but is not endorsed or certified by TMDB.

Wikipedia episode summaries (CC BY-SA 4.0). Rotten Tomatoes and Metacritic scores were read from their public pages for non-commercial research. Charts: CC BY 4.0, from the repository.