In any scientific experiment, if you want to find out if something works you isolate it. You make sure that all other factors that may influence the outcome of the experiment are as predictable as possible. Thus any reaction to the experiment is a reaction to the one unknown factor you allow.
In web design terms, any user feedback on or statistics of the site as a whole should ideally reflect only on your new idea. All items other than your new idea should be well known, even a bit boring, so that your visitors will use them without thinking and won't comment on them.
Of course, web development is not an exact science--far from it. Truly scientific experiments cannot be performed on web sites, since even the simplest site contains far too many unknown factors.
