What’s the p-value on this bad boy? Also I wonder if they controlled for confounding factors. Ones I can think of are: “Thanks” more likely on emails containing requests, which inherently requires response if from colleague; personality could dictate which valediction you choose, which could also be reflected on how you wrote your email; whether “Best” and “Thanks” are used can be very cultural - eg everyone at Microsoft uses “Thanks” ONLY, and it would be very weird to see a “Best” and extremely rare to see a “Cheers”, those same communities may happen to have a larger response rate since some companies have better email culture, thus given their non-random sampling since they just used the responses from a small handful of online communities, that could bias results.