I hope that they normalized the data to account for the number of words spoken during each debate, rally, etc... This is important because of Zipf's law: https://en.wikipedia.org/wiki/Zipf%27s_law

If they didn't, the findings are invalid.