Adsense experiment sample size

How much traffic do I need to do experiments?

[See_It_Now]

I'm pretty new to the Adsense business. I've been building some sites and, while things are still crawling, I think I can see where traffic can grow to good levels. So far I have been putting ad units on the pages without any real scientific basis as to number, placement, format, or colors.

I would like to begin to do some experiments, but I have some questions before I start.

- How much traffic do I need on a page to do a valid experiment? I would think that I need 100's of page impressions before I draw firm conclusions about a particular arrangement.

- Should I use clicks or earnings as the measure of success? My earnings per click vary a lot, so a new ad might get more clicks but lower earnings. Should I ignore a low epc and assume it will eventually catch up, or is that a sign that the new ad will consistently earn less per click?

- A true scientist would vary everything - number of units, placement, format, and color - independently, but that would make my experiments take forever. What do you suggest?

Thanks

[Hobbs]

a) Read well what others here write and start where they ended
b) Blend well
c) Start with the least number od ad units per page do not fill it up
d) The more traffic the more relevant your conclusions will be
e) Each test should not be less than one week
f) Document well all that you do, don't be afraid to backtrack and confirm previous results

[mzanzig]

g) Do only one change at a time (ideally the most promising one)

[jomaxx]

Hundreds of impressions is not very much. I would say you need somewhere in the hundreds of clicks in order to make the obvious choices, and thousands of clicks to make the subtler choices.