Not a very serious study. If he allowed forms to be processed even if captcha failed, he could of known how many of the failures were indeed legit clients. This

Not a very serious study. If he allowed forms to be processed even if captcha failed, he could of known how many of the failures were indeed legit clients. This way, it might of just been bots and the 2-3% change just a normal deviation.