Author here: we scraped every script website on the Internet. First we tried to normalize the dataset but only doing stats on the top 1,000 box office, but we were missing too many scripts. So we decided to go big and then display a cut of the data that's only films in the top 2,500 box office (we had about half of those).

We're aware of sampling error and the potential for cherry-picking, but also struggled to figure out what was a representative sample.