4 ms·
Unfortunately, that self-experiment has the problem of the subject (yourself) knowing which "treatment" you're getting. I'm not sure how to double-blind somethi
by scott_s 11y ago
Unfortunately, that self-experiment has the problem of the subject (yourself) knowing which "treatment" you're getting. I'm not sure how to double-blind something like this, though. It may require a study where we record subjects' sleep patters, make them use a screen at night, but don't tell them that it's the screen colors that are under experiment.
- gwern 11y agoEmpirically, I didn't notice and that almost screwed up the experiment; see the part where a driver change broke Redshift and I didn't notice for a month. (As well, I would note that you may be overestimating placebo or expectancy effects here. In the meta-analyses I've seen, randomization is much more important than placebo controls.)
- scott_s 11y agoInteresting - if I understand correctly, you're saying that you didn't notice for a month that no color balance was happening. Then perhaps an even simpler experiment preliminary experiment could be done to just see how well subjects are able to guess if a screen has been color balanced at night. If most people can't tell most of the time, then that simplifies doing an efficacy experiment.
- gwern 11y agoNot quite. The reddening is blatant, so if someone had asked me on any particular day if Redshift was on or not, I could always correctly answer. But it was randomized on a daily basis, so it is possible, even probable, for occasional long sequences like 100001 to happen where the shifting was on or off for multiple days in a row. What I'm saying is that I turned out to be paying so little attention to whether the shifting was on or off, certain that if it was off on a particular day that just meant the experiment was going as planned, that I didn't notice even 20+ days of a single setting.