I add this because no one seems to have mentioned it yet. Without experimenter freedom it is hard to make any inferences from the results of experiment. If the experimental choices you make are correlated with the stuff you're experimenting on in hard to understand ways then the experimental data you gather is essentially meaningless.

To make a dumb example, you measure some system and always observe outcome A. You might conclude that A is a property of the system you measured, but it just so happens that your decision to do the experiment is correlated with the system, so it always has A when you measure it, but it has other properties at other times.