p-hacking (and other data manipulations) made mainstream?

I am not a big fan of the last part of the talk that is too optimistic about randomized control trials and may make then look like some sort of panacea. RCTs have important limitations, especially when it comes to external validity. If you run enough of them, you are bound to find the kind of patterns Laura Arnold seems to criticizes using the 15 ingredients study (7:20 in the video). What works here now might not work there tomorrow. Eventually, you need structural models and sound theory to understand when a treatment works, and when it does not.

That’s the curse of the TED format. Arnold is somewhat critical of the format, but she has to fit in it. I am sure she and the organizations she advocates for know of the limits of RCTs. But there’s only so much you can say in 18 minutes if you want to conclude on an uplifting note.

Still, the overall message is very much worth spreading. It’s important to popularize notions like p-hacking and file drawer effect that rarely even make it into introductory statistics class. And it’s nice to see TED being a little critical of itself (or rather TEDx being critical of TED).