You would have to assume that I was subconsciously also lying in the predictions, to hide my foreknowledge, and how would I be able to precisely affect the sleep or Mnemosyne scores?
> Also, given that the period of the experiment is exactly a week, I also would control for 'LSD is the first thing exactly as often as the test"
? 3-day blocks imply a 6-day cycle, not 'exactly a week'. And I'm not sure what you mean by 'control' either.
And control: see http://thesocietypages.org/socimages/2013/01/12/what-does-it...
Yes, I'm aware of the statistical concept, and do controlling in a few other essays. My point was: 'control' is not an instantly obvious application. What am I controlling and how? Am I modeling it as a separate additive categorical variable? Quantitative variable? An interaction? (All different in R's formula notation, because they're different - +, *, :, and |) When you say
> I also would control for 'LSD is the first thing exactly as often as the test"
I don't even know what I would be controlling here, since I don't know what the order is supposed, what the 'test' is, and so on. You may think your suggestion is clear, but to me it's as clear as mud.
One way to do that is to make 2N packages containing a spiked and a non-spiked sample, with one sample marked 'take me first', and, in N cases, have the LSD-spiked sample be that sample. Of course, if you do that, it is even more important to not inform the subject about the history of he experiment before it is over.
I see. So you were suggesting throwing in a categorical variable for the day of week and including that as a additive covariate to control for fixed-effects on days. I haven't actually run into any of my experiments depending on day of week to any important degree, but I still try to avoid 7-day blocks to make that less of an issue. It's not like there's usually an advantage to running in a weekly block, so better safe than sorry.
Secondly, he may have wanted a null result.
Flaws can go in any direction, and make a design conservative or optimistic. In this case, the identified flaw does not undermine the actual result of the experiment - though it would have undermined the opposite (hypothetical) result of positive effects.
"This suggests the following design: a randomized dose on day 1, followed by days 2 and 3 off, then on day 4, drinking the second container; on day 7, examine the containers recording whether active/placebo and finally, starting over as day 1 of a new pair of 3-day blocks."
That means that, during the experiment, the subject can learn correlations between the test and the control ("So, I tend to go to the bathroom in the middle of the night in the test.")
If the subject has some (conscious or unconscious) prejudice for the result, he can use that information to adjust his self-reporting (again, consciously or unconsciously). For example, a subject may report a silly thought as a creative one instead of as a stupid one.
Also, given that the period of the experiment is exactly a week, I also would control for 'LSD is the first thing exactly as often as the test"