I was watching Back to the Future the other night -- it still holds up! I noticed that Michael J. Fox is pretty short and wondered if that could be because he started as a child actor. Older children are easier to handle, so picking a short young looking 10 year old to play an 8 year old makes sense. Presumably on average those short 10 year olds grow up to be short adults.
Fox got his first professional role at 15, playing a ten-year-old in Leo and Me, so that seemed promising. Also, turns out in California a six to eight-year-old can work four hours, while a child under six gets three under the rules for child actors.
Seems like an interesting enough theory to see if we could test it. So I fired up goose and asked it to look into it and went back to the movie.
While I watched the rest, the AI combined the height field in Wikidata with the IMDb non-commercial datasets. It found US actors born from 1950 through 1980, worked out the year of their first principal acting credit and compared that with their adult height.

There were 1,200 actors with usable gender data. The 131 who started before 18 were shorter: three centimetres for women and almost four for men. After adjusting for gender and birth year, the difference was 3.1 centimetres. Restricting the sample to people with at least twenty acting credits barely changed it.
That showed the correlation but not the explanation. The Wikidata sample is hardly a census, IMDb's first recorded credit is not always an actor's actual first job and maybe short people enter acting earlier for some other reason, maybe they are picked upon at school and so join the drama kids -- if the casting theory was right, child actors should also be older than the characters they play.
So I asked the AI to start with 1,000 actors who definitely had an English Wikipedia biography. Twenty agents each took a batch, read the biographies, followed links to early productions and checked up to five childhood roles.
They still needed watching. One result said a 13-year-old had played a four-year-old. I asked whether the production had used different actors for the different ages. It had: there were separate actors at four, eight and eleven, and the agent had assigned all three ages to one of them. In the final sample, 55 IMDb records were bad matches or otherwise unusable and I excluded another 33 suspicious results.
Character ages are rarely stated as numbers. I ended up with 208 roles where I could get an exact age for both actor and character. In 134 the actor was older, in 47 they were about the same age and in 27 the actor was younger. The median difference was one year. Looking only at the 114 cases where this was the actor's earliest role gave almost the same pattern.

The 208 roles are not a random sample. Character ages are more likely to be stated when the age matters to the story. Also, the actor's age was often calculated at release or broadcast rather than when filming started. That delay could account for the median difference of one year. So this does not prove why child actors are shorter, but actors were still much more often older than their characters than younger.
Here are the height data, actor-character pairs and research workbook.
So this amateur science ano 2026. The data is messy and there's no reviewer so hardly real science. But it is two experiments with the data attached, all from a question I asked halfway through a movie and reproduceable by anybody, especially anybody with access to an agent.
Of what you see in books, believe 75%. Of newspapers, believe 50%. And of TV news, believe 25% -- make that 5% if the anchorman wears a blazer.