3 ms·
> It's very unlikely that there exists a treatment where all the effect is on a 35-variable interaction and none of the effect is visible when looking at each v
by schoen 11y ago
> It's very unlikely that there exists a treatment where all the effect is on a 35-variable interaction and none of the effect is visible when looking at each variable singly.
Maybe something about genetic modifications where 35 different point modifications to the genome have to happen in order to allow synthesis of a working protein? If you measured the effect of each point modification, you might measure it as zero.
I'm analogizing from trying to do statistical tests of whether a password is correct. At least in systems that aren't vulnerable to timing attacks, you can't just test part of the password and see if you get a little bit more access to the system as a result. :-)