# How to determine multivariate confidence in laymans terms

I am trying to determine colors to best grab a web-users attention.

I have a simple system which randomly chooses one of 4 foreground colors, and one of 6 background colors.

I then save the stats in a table like this:

fore_color     back_color    shown    clicked
----------     ----------    -----    -------
red            green         16       1
blue           green         15       2


At the moment, I require every combination to be shown 250 times before I come to a conclusion about which combination is best... For a test size of ~6000 views.

I would like to know how I can reduce the size of the test while maintaining high confidence that we are making the correct choice (>90% admittedly, I don't know if my current method meets that standard).

Because I am only interested in 90% confidence for the best result, and not at all interested in stats for the losers, we can obviously disqualify early-bad-performers more quickly than final-contenders.

To be clear, my goal is to attain 90% confidence of the best results while showing each combination the least amount of times possible. I will be running this test over 1000 times on many different web pages, and would like to eventually include more variables (like font size).

If the math is simple, I could implement this directly in my own programing. If it's more complex, I'd be more interested in a (preferably free) package that I could feed the data through, and return confidence levels that a particular combination is or is not our winner.