Basic Tutorial On Split-Testing"" by Michel Fortin
This is pulled from my post on CopywritersBoard.com.
A script (like SIMA or Split Test Gold, etc) serves 2 versions of the same page. The 2nd version should be a carbon copy of the first page (your control), but with one variable changed — perhaps a change in headline, price, graphic, copy, offer, guarantee, etc.
So page “a” may be your existing control while page “b” is exactly the same with one slight change. (Always test just one variable at a time.)
Scenario of “a” versus “b” split test.
The script serves one page, alternately, with each visitor. It measures the response rate per page. Say, page “a” gets 100 hits, 2 sales (2% response). Page “b” gets 100 hits, 3 sales (3% response). The difference between “a” and “b” gives me the change. It’s a 1% difference. Based on the original 2%, that’s a 50% increase.
Let me show you a more specific example.
First, you install the script. You then set your 2 pages. Page “a” is your current control. Page “b” is an exact carbon copy of page “a,” but with a slight change. One variable, whatever it may be. Now…
Page “a” 213 hits, 2 sales, 0.9% conversion
Page “b” 188 hits, 3 sales, 1.1% conversion
Increase from “a” to “b:” 1.1 – 0.9 = 0.2% difference
To calculate the increase 0.2% represents:
Multiply it by 100, like 0.2 x 100 = 20
20 divided by 0.9% = 22% increase
Thus, page “b” with the slight change represents a 22% increase over page “a.” Thus, 0.9% (page “a”) x 1.22 (22% increase) = 1.1% (page “b”). In other words, add 22% of 0.9% to 0.9%, which equals 1.1%.
Now, is this empirical? No.
The amount of traffic above is very small. This example should not be considered as across-the-board numbers. But, it gives you a good indicator. (Don’t forget that a modification to your control may decrease your response, too. That’s fine, since it gives you the knowledge of what DOESN’T work — and not repeat it.)
Here’s a great site that teaches it to you is http://testandtrack.com/.
Tags: 
