A book from Hollywood Metrics
What 74,571 films and 1,571 screenplays reveal about how movies actually work
This project spent eighteen months measuring screenplays against how the finished films were received. Twenty structural measurements, 1,571 produced films. Here is everything it found, drawn on the actual scale a correlation lives on rather than one cropped to make the bars look like something.
screenplay.correlation_matrix_shipped, models.r2_budget_only and models.boxoffice_model_r2.The book is what happens when a project publishes that instead of burying it: how the number was found, why an earlier version of it was wrong, what it costs to admit that, and what does predict a film’s performance once the screenplay has been ruled out.
Each of these is a chapter. Each one carries the sample it was computed over, the method, and the reason it might be wrong, because a number without those is a rumour with a decimal point.
Nothing in the text of a screenplay predicts how the film was received
The budget explains three times more than twenty measurements of the script
Six films in seven never report what they made
Counting the silent films as failures inverts the cheap end entirely
The prestige slate is the most reliably profitable class in the catalogue
Horror returns more per dollar than anything else that reports
Three quarters of the screenplay corpus read as having no dialogue at all
The most-taught number in screenwriting splits 41 to 28 inside one corpus
One filter removed the majority of the source rows, unevenly
The bug was found by the people who wrote it. Then it was not fixed for months, and the numbers it produced were left standing.
Chapter nineteen
All three were measured for this book. None of them is reproduced in it: not a critic’s score against a named film, not a line of screenplay, not a page of anyone else’s writing.
74,571
A century of budgets, grosses, ratings and awards, filtered by one vote threshold that Part One spends a chapter interrogating.
1,571
Produced films, each matched to its script and measured on twenty structural features, twice: once with a broken parser and once with a working one.
926
A million words on how to write, direct, shoot and act, audited for what they agree on, what they contradict, and what they never mention.
15.46%
Six films in seven never say what they made. Every profitability claim ever published describes the seventh, and this one says so.
A 6 x 9 in paperback, 366 pages, black and white on 60# uncoated cream, perfect bound with a matte laminate cover. It is set in the same three faces this site uses, and the reference section at the back runs to about a thousand rows, which is the part that needs paper.
Buying here rather than through a retailer removes the retailer’s cut, which is why the price is lower and why the ebook comes with it. That is the whole of the pitch for buying direct; the rest of this page is about the book.
Printed when you order it, on the cream stock this panel is the colour of. No retailer takes a cut, and the ebook comes with it.
| One copy | $19.99 |
|---|---|
| Two or more | $18.39−8% |
| Five or more | $16.39−18% |
| Ten or more, for a room or a class | $14.39−28% |
Postage is on us over $75.00.
Every copy is printed after the order is placed, so the window above covers making it as well as posting it. 30 days to send it back.
No screenplay is quoted. No film is listed beside somebody else’s rating. Three corpora measured, none reproduced.
Appendix C
The argument is free. The thousand rows of reference tables are what the two hundred printed pages are for.
The full correlation matrix: twenty measurements against six outcomes, the rule behind it, and the questions it gets asked.
Every chapter with the idea it carries, the twenty features defined exactly, and the book's own statement of what it cannot establish.
What a copy costs at every quantity, what postage is to each zone, how long it takes and how to send it back.