I've been writing drinking game rules for the past few primary debates to, uh, encourage political engagement among my friends. The rules are basically just a list of each candidate's catchphrases, but I wanted try a more scientific approach for the upcoming debate: can data science create the ultimate drinking game?

Since June, there have been 7 debates, giving us hours of candidate dialogue. I wrote some quick code to scrape the debate transcripts and grab every single word uttered on stage - all 178,692 of them! I then used tf-idf to highlight three word phrases said often by each politician but rarely by the others. tf-idf is a convenient measure here because it devalues generic phrases used often by everyone, like “when I'm president…” or “we need to…”, making the final results more unique to each candidate.

Below are the results for everyone who'll be in the next debate. Hover over any phrase to read it in context.

As a bonus, here are the sayings of some candidates we’ve seen debate at least thrice previously, but who won’t be on stage this month. Don’t forget to hover over any phrase to read how the candidate uses it in context.

Whoever wins the Democratic primary will eventually debate Trump, who's been hosting his own campaign rallies lately. I grabbed 15 transcripts and pulled a total of 160,443 words from his campaign speeches. I then used the same tf-idf algorithm to surface the three-word-phrases he often uses, compared to any of the Democratic candidates.

I also grabbed Trump's dialogue from the 2015-16 primary debates and compared it to other Republicans in those same debates to show how his campaign rhetoric already differs from four years ago.

Disclaimer: as this is a study of language and rhetoric, I'm including direct quotes pulled from his transcripts. Transcripts don't fact-check. Take everything with a grain of salt.